Publications
Search
1.
Barker, Sarah Alicia; Miles-Novelo, Andreas J; Rizzo, Albert; Tuma, Regina M
Efficacy of Virtual Reality–Based Mindfulness Interventions: Systematic Review and Meta-Analysis Journal Article
In: JMIR XR Spatial Comput, vol. 3, pp. e90003–e90003, 2026, ISSN: 2818-3045.
@article{barker_efficacy_2026,
title = {Efficacy of Virtual Reality–Based Mindfulness Interventions: Systematic Review and Meta-Analysis},
author = {Sarah Alicia Barker and Andreas J Miles-Novelo and Albert Rizzo and Regina M Tuma},
url = {https://xr.jmir.org/2026/1/e90003},
doi = {10.2196/90003},
issn = {2818-3045},
year = {2026},
date = {2026-06-01},
urldate = {2026-07-09},
journal = {JMIR XR Spatial Comput},
volume = {3},
pages = {e90003–e90003},
abstract = {Background: Virtual reality–based mindfulness interventions (VRbMIs) increasingly populate studies as scalable tools for
stress and emotion regulation. However, findings across psychological outcomes are heterogeneous, and methodological
variation in intervention design, outcome measurement, and reporting practices limits cross-study comparability and cumulative synthesis.
Objective: A PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020–compliant systematic
review and meta-analysis was conducted to evaluate the psychological effects of VRbMIs published between 2023 and 2025,
examining methodological quality and outcome consistency across diverse study designs.
Methods: A PRISMA 2020–compliant systematic search was conducted across PubMed, PsycINFO, Scopus, Semantic
Scholar, and CORE from January 2023 to April 2025. Eligible studies included quantitative VRbMIs reporting psychological
outcomes. We assessed risk of bias using the Mixed Methods Appraisal Tool and conducted random-effects meta-analyses
using Hedges g where data were sufficient.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Background: Virtual reality–based mindfulness interventions (VRbMIs) increasingly populate studies as scalable tools for
stress and emotion regulation. However, findings across psychological outcomes are heterogeneous, and methodological
variation in intervention design, outcome measurement, and reporting practices limits cross-study comparability and cumulative synthesis.
Objective: A PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020–compliant systematic
review and meta-analysis was conducted to evaluate the psychological effects of VRbMIs published between 2023 and 2025,
examining methodological quality and outcome consistency across diverse study designs.
Methods: A PRISMA 2020–compliant systematic search was conducted across PubMed, PsycINFO, Scopus, Semantic
Scholar, and CORE from January 2023 to April 2025. Eligible studies included quantitative VRbMIs reporting psychological
outcomes. We assessed risk of bias using the Mixed Methods Appraisal Tool and conducted random-effects meta-analyses
using Hedges g where data were sufficient.
stress and emotion regulation. However, findings across psychological outcomes are heterogeneous, and methodological
variation in intervention design, outcome measurement, and reporting practices limits cross-study comparability and cumulative synthesis.
Objective: A PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020–compliant systematic
review and meta-analysis was conducted to evaluate the psychological effects of VRbMIs published between 2023 and 2025,
examining methodological quality and outcome consistency across diverse study designs.
Methods: A PRISMA 2020–compliant systematic search was conducted across PubMed, PsycINFO, Scopus, Semantic
Scholar, and CORE from January 2023 to April 2025. Eligible studies included quantitative VRbMIs reporting psychological
outcomes. We assessed risk of bias using the Mixed Methods Appraisal Tool and conducted random-effects meta-analyses
using Hedges g where data were sufficient.
2.
Zang, Tao; Awada, Mohamad; Becerik-Gerber, Burcin; Roll, Shawn C.; Lucas, Gale; Narayanan, Shrikanth
Does your office make you feel worse? Linking environmental exposure and temporal factors to cognitive symptoms Journal Article
In: Indoor Environments, vol. 3, no. 2, pp. 100171, 2026, ISSN: 29503620.
@article{zang_does_2026,
title = {Does your office make you feel worse? Linking environmental exposure and temporal factors to cognitive symptoms},
author = {Tao Zang and Mohamad Awada and Burcin Becerik-Gerber and Shawn C. Roll and Gale Lucas and Shrikanth Narayanan},
url = {https://linkinghub.elsevier.com/retrieve/pii/S2950362026000238},
doi = {10.1016/j.indenv.2026.100171},
issn = {29503620},
year = {2026},
date = {2026-06-01},
urldate = {2026-06-18},
journal = {Indoor Environments},
volume = {3},
number = {2},
pages = {100171},
abstract = {Symptoms such as fatigue, headache, low motivation, and difficulty concentrating affect office workers’ cognitive and overall well-being. This study investigates the dynamic relationships among temporal patterns, indoor environmental quality (IEQ) factors, and cognitive symptoms in real offices. We collected about four months of continuous environmental data and ecological momentary assessments (EMAs) from ten office workers and integrated XGBoost and SHAP analysis to identify those relationships. Model performance varied by symptom and participant. Under random train-test splitting, accuracy generally exceeded 70% with F1 scores ranging from 40% to 80%. Performance declined under a temporal split, highlighting the real-world challenge of predicting future symptom states from past observations. Based on SHAP analysis, hour of day and day of week emerged as significant predictors, revealing temporal patterns in cognitive symptoms. Within observed one-hour windows prior to reported symptoms, higher relative humidity was generally associated with increased risk of cognitive symptoms, while higher illuminance was associated with reduced symptom risk. Associations with sound pressure level and TVOCs varied across participants, with no consistent directions. Our data-driven approach provides insights into temporal factors and environmental exposures associated with cognitive symptoms using naturalistic data. These findings highlight the combined influence of temporal and IEQ factors on cognitive symptoms and contribute to the development of adaptive, user-centric workplaces that support cognitive well-being management.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Symptoms such as fatigue, headache, low motivation, and difficulty concentrating affect office workers’ cognitive and overall well-being. This study investigates the dynamic relationships among temporal patterns, indoor environmental quality (IEQ) factors, and cognitive symptoms in real offices. We collected about four months of continuous environmental data and ecological momentary assessments (EMAs) from ten office workers and integrated XGBoost and SHAP analysis to identify those relationships. Model performance varied by symptom and participant. Under random train-test splitting, accuracy generally exceeded 70% with F1 scores ranging from 40% to 80%. Performance declined under a temporal split, highlighting the real-world challenge of predicting future symptom states from past observations. Based on SHAP analysis, hour of day and day of week emerged as significant predictors, revealing temporal patterns in cognitive symptoms. Within observed one-hour windows prior to reported symptoms, higher relative humidity was generally associated with increased risk of cognitive symptoms, while higher illuminance was associated with reduced symptom risk. Associations with sound pressure level and TVOCs varied across participants, with no consistent directions. Our data-driven approach provides insights into temporal factors and environmental exposures associated with cognitive symptoms using naturalistic data. These findings highlight the combined influence of temporal and IEQ factors on cognitive symptoms and contribute to the development of adaptive, user-centric workplaces that support cognitive well-being management.
3.
Cai, Yunxuan; Xiang, Sitao; Li, Zongjian; Chen, Haiwei; Zhao, Yajie
Bringing Diversity from Diffusion Models to Semantic-Guided Face Asset Generation Journal Article
In: ACM Trans. Graph., vol. 45, no. 3, pp. 1–19, 2026, ISSN: 0730-0301, 1557-7368.
@article{cai_bringing_2026,
title = {Bringing Diversity from Diffusion Models to Semantic-Guided Face Asset Generation},
author = {Yunxuan Cai and Sitao Xiang and Zongjian Li and Haiwei Chen and Yajie Zhao},
url = {https://dl.acm.org/doi/10.1145/3793859},
doi = {10.1145/3793859},
issn = {0730-0301, 1557-7368},
year = {2026},
date = {2026-06-01},
urldate = {2026-05-21},
journal = {ACM Trans. Graph.},
volume = {45},
number = {3},
pages = {1–19},
abstract = {High-quality 3D face asset creation remains costly due to reliance on controlled capture setups and manual processing, limiting scalability and diversity. We introduce a fully automated, semantically controllable framework for generating PBR-ready 3D facial assets without requiring dedicated scans. Our pipeline begins with a diffusion-based data synthesis stage, where 2D portrait samples from a pre-trained diffusion model are converted into 44K textured 3D face reconstructions via our proposed geometry recovery and texture normalization algorithm, which aligns arbitrarily shaded outputs into clean albedo space. Using this dataset, we train a disentangled adversarial generator that maps semantic attributes (age, gender, ethnicity) to UV-space geometry and albedo, enabling both direct sampling and continuous latent editing while preserving identity. A refinement stage further produces PBR materials and secondary assets (eyeballs, teeth, gums). The resulting system supports controllable face generation and post-editing in real time and exports directly to standard rendering and animation pipelines. We evaluate each component extensively and provide a web-based interactive interface to showcase practical deployment.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
High-quality 3D face asset creation remains costly due to reliance on controlled capture setups and manual processing, limiting scalability and diversity. We introduce a fully automated, semantically controllable framework for generating PBR-ready 3D facial assets without requiring dedicated scans. Our pipeline begins with a diffusion-based data synthesis stage, where 2D portrait samples from a pre-trained diffusion model are converted into 44K textured 3D face reconstructions via our proposed geometry recovery and texture normalization algorithm, which aligns arbitrarily shaded outputs into clean albedo space. Using this dataset, we train a disentangled adversarial generator that maps semantic attributes (age, gender, ethnicity) to UV-space geometry and albedo, enabling both direct sampling and continuous latent editing while preserving identity. A refinement stage further produces PBR materials and secondary assets (eyeballs, teeth, gums). The resulting system supports controllable face generation and post-editing in real time and exports directly to standard rendering and animation pipelines. We evaluate each component extensively and provide a web-based interactive interface to showcase practical deployment.
4.
Curhan, Jared R.; Gratch, Jonathan
Introduction to Special Issue Artificial Intelligence and Negotiation: A Framework for an Emerging Field Journal Article
In: Negotiation Journal, vol. 42, 2026, ISSN: 0748-4526, 1571-9979.
@article{curhan_introduction_2026,
title = {Introduction to Special Issue Artificial Intelligence and Negotiation: A Framework for an Emerging Field},
author = {Jared R. Curhan and Jonathan Gratch},
url = {https://direct.mit.edu/ngtn/article/doi/10.1162/NGTN.a.69/136835/Introduction-to-Special-Issue-Artificial},
doi = {10.1162/NGTN.a.69},
issn = {0748-4526, 1571-9979},
year = {2026},
date = {2026-05-01},
urldate = {2026-06-16},
journal = {Negotiation Journal},
volume = {42},
abstract = {Abstract
Artificial intelligence is increasingly embedded in the practice, study, and teaching of negotiation, creating the need for a clear conceptual vocabulary for this emerging field. In this article, we propose a spatial and functional framework, identifying four primary roles AI may occupy in negotiation: as agent in front of the negotiator, as coach beside the negotiator, as backtable support behind the negotiator, and as mediator or facilitator between the parties. We also identify three broader domains in which AI is transforming the negotiation field: teaching, research, and real-world implementation. Although these categories are analytically distinct, they often overlap in practice, especially in field settings where multiple AI functions are integrated within the same system. By organizing these roles and domains within a single framework, we seek to clarify how AI is revolutionizing negotiation, and how its various applications relate to one another.
The significance of AI in negotiation lies not only in what these systems can do, but in how they are designed and positioned relative to human judgment, institutional context, and the conduct of negotiation itself. Drawing on the contributions in this special issue, we examine how AI is reshaping processes, expanding access to training, reorienting empirical research, and supporting practitioners in complex environments. At the same time, these developments raise enduring questions about delegation, transparency, equity, accountability, and the preservation of human judgment. We argue that the most promising trajectory for AI in negotiation is one of complementarity: systems designed to augment human capabilities, preserve meaningful human involvement, and strengthen negotiation practice rather than displace the human capacities that give it meaning. By clarifying roles and articulating normative principles for design and use, this framework offers both an analytic foundation for the field and guidance for its responsible development.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Abstract
Artificial intelligence is increasingly embedded in the practice, study, and teaching of negotiation, creating the need for a clear conceptual vocabulary for this emerging field. In this article, we propose a spatial and functional framework, identifying four primary roles AI may occupy in negotiation: as agent in front of the negotiator, as coach beside the negotiator, as backtable support behind the negotiator, and as mediator or facilitator between the parties. We also identify three broader domains in which AI is transforming the negotiation field: teaching, research, and real-world implementation. Although these categories are analytically distinct, they often overlap in practice, especially in field settings where multiple AI functions are integrated within the same system. By organizing these roles and domains within a single framework, we seek to clarify how AI is revolutionizing negotiation, and how its various applications relate to one another.
The significance of AI in negotiation lies not only in what these systems can do, but in how they are designed and positioned relative to human judgment, institutional context, and the conduct of negotiation itself. Drawing on the contributions in this special issue, we examine how AI is reshaping processes, expanding access to training, reorienting empirical research, and supporting practitioners in complex environments. At the same time, these developments raise enduring questions about delegation, transparency, equity, accountability, and the preservation of human judgment. We argue that the most promising trajectory for AI in negotiation is one of complementarity: systems designed to augment human capabilities, preserve meaningful human involvement, and strengthen negotiation practice rather than displace the human capacities that give it meaning. By clarifying roles and articulating normative principles for design and use, this framework offers both an analytic foundation for the field and guidance for its responsible development.
Artificial intelligence is increasingly embedded in the practice, study, and teaching of negotiation, creating the need for a clear conceptual vocabulary for this emerging field. In this article, we propose a spatial and functional framework, identifying four primary roles AI may occupy in negotiation: as agent in front of the negotiator, as coach beside the negotiator, as backtable support behind the negotiator, and as mediator or facilitator between the parties. We also identify three broader domains in which AI is transforming the negotiation field: teaching, research, and real-world implementation. Although these categories are analytically distinct, they often overlap in practice, especially in field settings where multiple AI functions are integrated within the same system. By organizing these roles and domains within a single framework, we seek to clarify how AI is revolutionizing negotiation, and how its various applications relate to one another.
The significance of AI in negotiation lies not only in what these systems can do, but in how they are designed and positioned relative to human judgment, institutional context, and the conduct of negotiation itself. Drawing on the contributions in this special issue, we examine how AI is reshaping processes, expanding access to training, reorienting empirical research, and supporting practitioners in complex environments. At the same time, these developments raise enduring questions about delegation, transparency, equity, accountability, and the preservation of human judgment. We argue that the most promising trajectory for AI in negotiation is one of complementarity: systems designed to augment human capabilities, preserve meaningful human involvement, and strengthen negotiation practice rather than displace the human capacities that give it meaning. By clarifying roles and articulating normative principles for design and use, this framework offers both an analytic foundation for the field and guidance for its responsible development.
5.
Han, Zicheng; Becerik-Gerber, Burcin; Liu, Ruying; Lucas, Gale; Doolab, Yogin; Zein, Sara; Atkins, Margo; Blass, Deborah; Miles, Lachlan
Emergency dynamics modeling with an agent-based framework: Active shooter scenario case study Journal Article
In: International Journal of Disaster Risk Reduction, vol. 142, pp. 106226, 2026, ISSN: 22124209.
@article{han_emergency_2026,
title = {Emergency dynamics modeling with an agent-based framework: Active shooter scenario case study},
author = {Zicheng Han and Burcin Becerik-Gerber and Ruying Liu and Gale Lucas and Yogin Doolab and Sara Zein and Margo Atkins and Deborah Blass and Lachlan Miles},
url = {https://linkinghub.elsevier.com/retrieve/pii/S2212420926002384},
doi = {10.1016/j.ijdrr.2026.106226},
issn = {22124209},
year = {2026},
date = {2026-05-01},
urldate = {2026-06-16},
journal = {International Journal of Disaster Risk Reduction},
volume = {142},
pages = {106226},
abstract = {Modeling human-induced emergencies requires a structured representation of interactions among occupants, hazards, and built environments. While existing simulation approaches provide important insights, behavioral processes, hazard progression, and architectural conditions are often examined separately. This study develops a parameterized agent-based modeling framework that integrates heterogeneous occupant behaviors, hazard dynamics, and environmental configurations within a commercially available crowd-simulation platform to extend its usability within the academic and professional communities. The framework comprises environmental modeling, agent definition, interaction logic, risk evaluation, and simulation execution, enabling systematic configuration of behavioral rules and scenario parameters through customized scripting. An Active Shooter Incident (ASI) in a school building is simulated as a demonstrative case study to illustrate the framework's capability to configure and test emergency scenarios under different behavioral, hazard and environmental assumptions. Using the framework, the study examines the effects of classroom occupancy, door width, and door-locking mechanisms on occupant safety, evaluated using a token-based high-risk metric and factorial statistical analysis. Results indicate that higher occupancy increases exposure risk, timely door locking substantially improves safety outcomes, and the influence of door width is context-dependent. The consistency of these patterns with prior research supports the behavioral plausibility of the modeling approach. Overall, the study contributes a practical and extensible framework for operationalizing human–hazard–environment interactions in similar agent-based simulations and for generating scenario-specific insights to support design and preparedness strategies in human-induced emergency contexts.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Modeling human-induced emergencies requires a structured representation of interactions among occupants, hazards, and built environments. While existing simulation approaches provide important insights, behavioral processes, hazard progression, and architectural conditions are often examined separately. This study develops a parameterized agent-based modeling framework that integrates heterogeneous occupant behaviors, hazard dynamics, and environmental configurations within a commercially available crowd-simulation platform to extend its usability within the academic and professional communities. The framework comprises environmental modeling, agent definition, interaction logic, risk evaluation, and simulation execution, enabling systematic configuration of behavioral rules and scenario parameters through customized scripting. An Active Shooter Incident (ASI) in a school building is simulated as a demonstrative case study to illustrate the framework's capability to configure and test emergency scenarios under different behavioral, hazard and environmental assumptions. Using the framework, the study examines the effects of classroom occupancy, door width, and door-locking mechanisms on occupant safety, evaluated using a token-based high-risk metric and factorial statistical analysis. Results indicate that higher occupancy increases exposure risk, timely door locking substantially improves safety outcomes, and the influence of door width is context-dependent. The consistency of these patterns with prior research supports the behavioral plausibility of the modeling approach. Overall, the study contributes a practical and extensible framework for operationalizing human–hazard–environment interactions in similar agent-based simulations and for generating scenario-specific insights to support design and preparedness strategies in human-induced emergency contexts.
6.
Gratch, Jonathan
Agents at the Table: Balancing Strategy, Participation, and Cooperation in the Design of AI Negotiators and Mediators Journal Article
In: Negotiation Journal, vol. 42, 2026, ISSN: 0748-4526, 1571-9979.
@article{gratch_agents_2026,
title = {Agents at the Table: Balancing Strategy, Participation, and Cooperation in the Design of AI Negotiators and Mediators},
author = {Jonathan Gratch},
url = {https://direct.mit.edu/ngtn/article/doi/10.1162/NGTN.a.65/136731/Agents-at-the-Table-Balancing-Strategy},
doi = {10.1162/NGTN.a.65},
issn = {0748-4526, 1571-9979},
year = {2026},
date = {2026-05-01},
urldate = {2026-06-16},
journal = {Negotiation Journal},
volume = {42},
abstract = {Abstract
Emerging advances in artificial intelligence—especially large language models—raise the prospect of AI systems acting as full participants in negotiation rather than merely supporting tools. This article surveys the evolution of automated negotiators and mediators and outlines three foundational design choices that shape their effectiveness, and indeed, how “effectiveness” is defined: whether systems should follow rational decision-making or mirror human relational and emotional behavior; whether people should delegate negotiation to autonomous agents or participate in the negotiation process; and how systems should balance self-interest with joint gains. Although these choices were once treated as dichotomous, recent systems increasingly integrate elements of each. These developments aim to align strategic decision-making with relational communication in ways that human counterparts judge as procedurally fair. The article illustrates these trends through four contributions presented at the 2025 AI Negotiation Summit at MIT.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Abstract
Emerging advances in artificial intelligence—especially large language models—raise the prospect of AI systems acting as full participants in negotiation rather than merely supporting tools. This article surveys the evolution of automated negotiators and mediators and outlines three foundational design choices that shape their effectiveness, and indeed, how “effectiveness” is defined: whether systems should follow rational decision-making or mirror human relational and emotional behavior; whether people should delegate negotiation to autonomous agents or participate in the negotiation process; and how systems should balance self-interest with joint gains. Although these choices were once treated as dichotomous, recent systems increasingly integrate elements of each. These developments aim to align strategic decision-making with relational communication in ways that human counterparts judge as procedurally fair. The article illustrates these trends through four contributions presented at the 2025 AI Negotiation Summit at MIT.
Emerging advances in artificial intelligence—especially large language models—raise the prospect of AI systems acting as full participants in negotiation rather than merely supporting tools. This article surveys the evolution of automated negotiators and mediators and outlines three foundational design choices that shape their effectiveness, and indeed, how “effectiveness” is defined: whether systems should follow rational decision-making or mirror human relational and emotional behavior; whether people should delegate negotiation to autonomous agents or participate in the negotiation process; and how systems should balance self-interest with joint gains. Although these choices were once treated as dichotomous, recent systems increasingly integrate elements of each. These developments aim to align strategic decision-making with relational communication in ways that human counterparts judge as procedurally fair. The article illustrates these trends through four contributions presented at the 2025 AI Negotiation Summit at MIT.
7.
Hans, Soham; Wang, Yunzhe; Ustun, Volkan
Egocentric Team AI: Enabling Tactical Reasoning from the Operator’s View Journal Article
In: AAAI-SS, vol. 8, no. 1, pp. 45–49, 2026, ISSN: 2994-4317.
@article{hans_egocentric_2026,
title = {Egocentric Team AI: Enabling Tactical Reasoning from the Operator’s View},
author = {Soham Hans and Yunzhe Wang and Volkan Ustun},
url = {https://ojs.aaai.org/index.php/AAAI-SS/article/view/42516},
doi = {10.1609/aaaiss.v8i1.42516},
issn = {2994-4317},
year = {2026},
date = {2026-05-01},
urldate = {2026-06-16},
journal = {AAAI-SS},
volume = {8},
number = {1},
pages = {45–49},
abstract = {The future of human–AI integration in high-stakes environments—ranging from defense operations to emergency response—requires AI systems that function as intuitive teammates rather than isolated tools. Yet a persistent perspective gap remains: video understanding models rely on third-person, broadcast-style views, while multi-agent reinforcement learning (MARL) systems operate on egocentric inputs but often depend on centralized critics during training, reducing the need for decentralized policies to internalize team-centric structure. As a result, agents may optimize behavior without learning policy-level representations of the tactical picture grounded in their own observations. To address this gap, we propose Egocentric Team AI, a research direction centered on learning implicit, distributed Common Operating Pictures from first-person views. Building on our prior work with the X-Ego-CS dataset and Cross-Ego Contrastive Learning (CECL), we outline how cross-egocentric representation alignment can be extended from passive video understanding to active multi-agent control. Specifically, we propose integrating cross-ego contrastive objectives into MARL within a multi-agent Doom-based environment as a testbed for decentralized, team-aware policy learning. By positioning cross-egocentric alignment as an inductive bias for decentralized coordination, this work charts a path toward embodied systems capable of adaptive cooperation without explicit communication.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The future of human–AI integration in high-stakes environments—ranging from defense operations to emergency response—requires AI systems that function as intuitive teammates rather than isolated tools. Yet a persistent perspective gap remains: video understanding models rely on third-person, broadcast-style views, while multi-agent reinforcement learning (MARL) systems operate on egocentric inputs but often depend on centralized critics during training, reducing the need for decentralized policies to internalize team-centric structure. As a result, agents may optimize behavior without learning policy-level representations of the tactical picture grounded in their own observations. To address this gap, we propose Egocentric Team AI, a research direction centered on learning implicit, distributed Common Operating Pictures from first-person views. Building on our prior work with the X-Ego-CS dataset and Cross-Ego Contrastive Learning (CECL), we outline how cross-egocentric representation alignment can be extended from passive video understanding to active multi-agent control. Specifically, we propose integrating cross-ego contrastive objectives into MARL within a multi-agent Doom-based environment as a testbed for decentralized, team-aware policy learning. By positioning cross-egocentric alignment as an inductive bias for decentralized coordination, this work charts a path toward embodied systems capable of adaptive cooperation without explicit communication.
8.
Mozgai, Sharon; Fast, Ed; Leeds, Andrew; Sookiassian, Edwin; Kim, Kevin; Hartholt, Arno
Rapid personalized doppelgänger avatar generation: Dyadic evaluation of the TAC-Twin virtual human pipeline. Proceedings Article
In: 2026.
@inproceedings{mozgai_rapid_2026,
title = {Rapid personalized doppelgänger avatar generation: Dyadic evaluation of the TAC-Twin virtual human pipeline.},
author = {Sharon Mozgai and Ed Fast and Andrew Leeds and Edwin Sookiassian and Kevin Kim and Arno Hartholt},
url = {https://openaccess.cms-conferences.org/publications/book/978-1-964867-77-9/article/978-1-964867-77-9_39},
doi = {10.54941/ahfe1007194},
year = {2026},
date = {2026-04-01},
urldate = {2026-06-18},
abstract = {Rapid generation of personalized doppelgänger avatars remains challenging due to the
time, expertise, and system integration typically required to produce virtual humans
suitable for interactive use. TAC-Twin is a modular avatar generation pipeline developed
within the Virtual Human Toolkit (VHToolkit) that converts a single photograph into a
fully rigged avatar deployable in Unity-based virtual human systems using productionready tools and an interoperable architecture. This paper presents an exploratory dyadic
evaluation of TAC-Twin to examine how the pipeline performs in practice when generating
self-, partner-, and generic avatars for a standardized scenario. Findings indicate
that the pipeline supports efficient, repeatable avatar creation and produces visually
recognizable avatars appropriate for research deployment. Participant feedback further
suggests that perceived avatar quality was shaped not only by facial resemblance, but
by how personalization carried through to facial animation, movement, and behavioral
coherence during scenario playback. The evaluation highlights that advances in avatar
generation alone are insufficient to sustain perceived realism over repeated exposure
if downstream expressive and behavioral layers are not aligned, underscoring the
importance of end-to-end pipeline design for personalized virtual humans.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Rapid generation of personalized doppelgänger avatars remains challenging due to the
time, expertise, and system integration typically required to produce virtual humans
suitable for interactive use. TAC-Twin is a modular avatar generation pipeline developed
within the Virtual Human Toolkit (VHToolkit) that converts a single photograph into a
fully rigged avatar deployable in Unity-based virtual human systems using productionready tools and an interoperable architecture. This paper presents an exploratory dyadic
evaluation of TAC-Twin to examine how the pipeline performs in practice when generating
self-, partner-, and generic avatars for a standardized scenario. Findings indicate
that the pipeline supports efficient, repeatable avatar creation and produces visually
recognizable avatars appropriate for research deployment. Participant feedback further
suggests that perceived avatar quality was shaped not only by facial resemblance, but
by how personalization carried through to facial animation, movement, and behavioral
coherence during scenario playback. The evaluation highlights that advances in avatar
generation alone are insufficient to sustain perceived realism over repeated exposure
if downstream expressive and behavioral layers are not aligned, underscoring the
importance of end-to-end pipeline design for personalized virtual humans.
time, expertise, and system integration typically required to produce virtual humans
suitable for interactive use. TAC-Twin is a modular avatar generation pipeline developed
within the Virtual Human Toolkit (VHToolkit) that converts a single photograph into a
fully rigged avatar deployable in Unity-based virtual human systems using productionready tools and an interoperable architecture. This paper presents an exploratory dyadic
evaluation of TAC-Twin to examine how the pipeline performs in practice when generating
self-, partner-, and generic avatars for a standardized scenario. Findings indicate
that the pipeline supports efficient, repeatable avatar creation and produces visually
recognizable avatars appropriate for research deployment. Participant feedback further
suggests that perceived avatar quality was shaped not only by facial resemblance, but
by how personalization carried through to facial animation, movement, and behavioral
coherence during scenario playback. The evaluation highlights that advances in avatar
generation alone are insufficient to sustain perceived realism over repeated exposure
if downstream expressive and behavioral layers are not aligned, underscoring the
importance of end-to-end pipeline design for personalized virtual humans.
9.
Wang, Yunzhe; Xu, Runhui; Zheng, Kexin; Zhang, Tianyi; Kogundi, Jayavibhav Niranjan; Hans, Soham; Ustun, Volkan
GameplayQA: A Benchmarking Framework for Decision-Dense POV-Synced Multi-Video Understanding of 3D Virtual Agents Miscellaneous
2026, (arXiv:2603.24329 [cs.CL]).
@misc{wang_gameplayqa_2026,
title = {GameplayQA: A Benchmarking Framework for Decision-Dense POV-Synced Multi-Video Understanding of 3D Virtual Agents},
author = {Yunzhe Wang and Runhui Xu and Kexin Zheng and Tianyi Zhang and Jayavibhav Niranjan Kogundi and Soham Hans and Volkan Ustun},
url = {http://arxiv.org/abs/2603.24329},
doi = {10.48550/arXiv.2603.24329},
year = {2026},
date = {2026-04-01},
urldate = {2026-06-18},
publisher = {arXiv},
abstract = {Multimodal LLMs are increasingly deployed as perceptual backbones for autonomous agents in 3D environments, from robotics to virtual worlds. These applications require agents to perceive rapid state changes, attribute actions to the correct entities, and reason about concurrent multi-agent behaviors from a first-person perspective, capabilities that existing benchmarks do not adequately evaluate. We introduce GameplayQA, a framework for evaluating agentic-centric perception and reasoning through video understanding. Specifically, we densely annotate multiplayer 3D gameplay videos at 1.22 labels/second, with time-synced, concurrent captions of states, actions, and events structured around a triadic system of Self, Other Agents, and the World, a natural decomposition for multi-agent environments. From these annotations, we refined 2.4K diagnostic QA pairs organized into three levels of cognitive complexity, accompanied by a structured distractor taxonomy that enables fine-grained analysis of where models hallucinate. Evaluation of frontier MLLMs reveals a substantial gap from human performance, with common failures in temporal and cross-video grounding, agent-role attribution, and handling the decision density of the game. We hope GameplayQA stimulates future research at the intersection of embodied AI, agentic perception, and world modeling.},
note = {arXiv:2603.24329 [cs.CL]},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Multimodal LLMs are increasingly deployed as perceptual backbones for autonomous agents in 3D environments, from robotics to virtual worlds. These applications require agents to perceive rapid state changes, attribute actions to the correct entities, and reason about concurrent multi-agent behaviors from a first-person perspective, capabilities that existing benchmarks do not adequately evaluate. We introduce GameplayQA, a framework for evaluating agentic-centric perception and reasoning through video understanding. Specifically, we densely annotate multiplayer 3D gameplay videos at 1.22 labels/second, with time-synced, concurrent captions of states, actions, and events structured around a triadic system of Self, Other Agents, and the World, a natural decomposition for multi-agent environments. From these annotations, we refined 2.4K diagnostic QA pairs organized into three levels of cognitive complexity, accompanied by a structured distractor taxonomy that enables fine-grained analysis of where models hallucinate. Evaluation of frontier MLLMs reveals a substantial gap from human performance, with common failures in temporal and cross-video grounding, agent-role attribution, and handling the decision density of the game. We hope GameplayQA stimulates future research at the intersection of embodied AI, agentic perception, and world modeling.
10.
Xiong, Xinmiao; Liu, Bangya; Wang, Hao; Li, Dayou; Chen, Nuo; Feng, Andrew; Ding, Mingyu; Banerjee, Suman; Zhou, Yang; Fan, Zhiwen
Accelerating Transformer-Based Monocular SLAM via Geometric Utility Scoring Miscellaneous
2026, (arXiv:2604.08718 [cs.CV]).
@misc{xiong_accelerating_2026,
title = {Accelerating Transformer-Based Monocular SLAM via Geometric Utility Scoring},
author = {Xinmiao Xiong and Bangya Liu and Hao Wang and Dayou Li and Nuo Chen and Andrew Feng and Mingyu Ding and Suman Banerjee and Yang Zhou and Zhiwen Fan},
url = {http://arxiv.org/abs/2604.08718},
doi = {10.48550/arXiv.2604.08718},
year = {2026},
date = {2026-04-01},
urldate = {2026-06-18},
publisher = {arXiv},
abstract = {Geometric Foundation Models (GFMs) have recently advanced monocular SLAM by providing robust, calibration-free 3D priors. However, deploying these models on dense video streams introduces significant computational redundancy. Current GFM-based SLAM systems typically rely on post hoc keyframe selection. Because of this, they must perform expensive dense geometric decoding simply to determine whether a frame contains novel geometry, resulting in late rejection and wasted computation. To mitigate this inefficiency, we propose LeanGate, a lightweight feed-forward frame-gating network. LeanGate predicts a geometric utility score to assess a frame's mapping value prior to the heavy GFM feature extraction and matching stages. As a predictive plug-and-play module, our approach bypasses over 90% of redundant frames. Evaluations on standard SLAM benchmarks demonstrate that LeanGate reduces tracking FLOPs by more than 85% and achieves a 5x end-to-end throughput speedup. Furthermore, it maintains the tracking and mapping accuracy of dense baselines. Project page: https://lean-gate.github.io/},
note = {arXiv:2604.08718 [cs.CV]},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Geometric Foundation Models (GFMs) have recently advanced monocular SLAM by providing robust, calibration-free 3D priors. However, deploying these models on dense video streams introduces significant computational redundancy. Current GFM-based SLAM systems typically rely on post hoc keyframe selection. Because of this, they must perform expensive dense geometric decoding simply to determine whether a frame contains novel geometry, resulting in late rejection and wasted computation. To mitigate this inefficiency, we propose LeanGate, a lightweight feed-forward frame-gating network. LeanGate predicts a geometric utility score to assess a frame's mapping value prior to the heavy GFM feature extraction and matching stages. As a predictive plug-and-play module, our approach bypasses over 90% of redundant frames. Evaluations on standard SLAM benchmarks demonstrate that LeanGate reduces tracking FLOPs by more than 85% and achieves a 5x end-to-end throughput speedup. Furthermore, it maintains the tracking and mapping accuracy of dense baselines. Project page: https://lean-gate.github.io/
11.
Feng, Andrew Zhuoer; Wang, Cunxiang; Luo, Yu; Fan, Lin; Zhou, Yilin; Wang, Zikang; Gu, Xiaotao; Tang, Jie; Wang, Hongning; Huang, Minlie
HoWToBench: Holistic Evaluation for LLM's Capability in Human-level Writing using Tree of Writing Miscellaneous
2026, (arXiv:2604.19071 [cs.CL]).
@misc{feng_howtobench_2026,
title = {HoWToBench: Holistic Evaluation for LLM's Capability in Human-level Writing using Tree of Writing},
author = {Andrew Zhuoer Feng and Cunxiang Wang and Yu Luo and Lin Fan and Yilin Zhou and Zikang Wang and Xiaotao Gu and Jie Tang and Hongning Wang and Minlie Huang},
url = {http://arxiv.org/abs/2604.19071},
doi = {10.48550/arXiv.2604.19071},
year = {2026},
date = {2026-04-01},
urldate = {2026-06-17},
publisher = {arXiv},
abstract = {Evaluating the writing capabilities of large language models (LLMs) remains a significant challenge due to the multidimensional nature of writing skills and the limitations of existing metrics. LLM's performance in thousand-words level and open-ended writing is inadequately assessed by traditional reference-based metrics or modern LLM-as-a-judge methods. We propose Tree-of-Writing (ToW), to resolve the implicit inconsistency often found when LLM-as-a-judge aggregates all sub-features in text evaluation. ToW incorporates a tree-structured workflow by explicitly modeling the aggregation weights of sub-features. We also present HowToBench, a large-scale Chinese writing benchmark encompassing 12 genres and 1302 instructions across three task categories: contextual completion, outline-guided writing, and open-ended generation. ToW successfully mitigates the biases, achieving a 0.93 Pearson correlation with human judgments. Furthermore, we detect that both overlap-based text generation metrics and popular LLM-as-a-judge practices are vulnerable to textual disturbances, while ToW is robust to them. We also uncover a negative correlation between input length and content-related scores in the Guide task, showcasing that it cannot be simply improved by input-side information piling.},
note = {arXiv:2604.19071 [cs.CL]},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Evaluating the writing capabilities of large language models (LLMs) remains a significant challenge due to the multidimensional nature of writing skills and the limitations of existing metrics. LLM's performance in thousand-words level and open-ended writing is inadequately assessed by traditional reference-based metrics or modern LLM-as-a-judge methods. We propose Tree-of-Writing (ToW), to resolve the implicit inconsistency often found when LLM-as-a-judge aggregates all sub-features in text evaluation. ToW incorporates a tree-structured workflow by explicitly modeling the aggregation weights of sub-features. We also present HowToBench, a large-scale Chinese writing benchmark encompassing 12 genres and 1302 instructions across three task categories: contextual completion, outline-guided writing, and open-ended generation. ToW successfully mitigates the biases, achieving a 0.93 Pearson correlation with human judgments. Furthermore, we detect that both overlap-based text generation metrics and popular LLM-as-a-judge practices are vulnerable to textual disturbances, while ToW is robust to them. We also uncover a negative correlation between input length and content-related scores in the Guide task, showcasing that it cannot be simply improved by input-side information piling.
12.
Melnychenko, Olga
HUMAN CAPITAL IN UKRAINE: PROBLEMS OF RECOVERY AND DEVELOPMENT (THE EDUCOLOGICAL ASPECTS) Journal Article
In: Continuing Professional Education Theory and Practice, pp. 7, 2026, ISSN: 1609-8595, 2412-0774.
@article{melnychenko_human_2026,
title = {HUMAN CAPITAL IN UKRAINE: PROBLEMS OF RECOVERY AND DEVELOPMENT (THE EDUCOLOGICAL ASPECTS)},
author = {Olga Melnychenko},
url = {https://npo.kubg.edu.ua/article/view/354893},
doi = {10.28925/2412-0774.2026.1.1},
issn = {1609-8595, 2412-0774},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-18},
journal = {Continuing Professional Education Theory and Practice},
pages = {7},
abstract = {The article examines the theoretical and practical foundations of human capital recovery as one of the key factors ensuring the economic and social development of Ukraine and highlights the role of education in this process. In the context of profound socio-economic transformations caused by the war, mass emigration, demographic losses, and the crisis of social institutions, the renewal and development of human capital have become issues of strategic importance. The author emphasizes that the loss of qualified professionals, forced migration, declining quality of education and healthcare, growing psychological exhaustion of the population, and the overall deterioration of the demographic situation significantly reduce the state’s capacity to respond effectively to emerging challenges and to sustain and renew human capital. The article argues that human capital should not be viewed solely as an economic category but rather as a multidimensional phenomenon that encompasses social, cultural, institutional, and value-based components. The relationship between the quality of human capital and the resilience of the entire social system at national, regional, and local community levels is analyzed. Given the contemporary challenges faced by Ukraine in wartime conditions, it is crucial to identify mechanisms that would enable the integration of human development policy into the national development strategy through a dedicated programme for the recovery and development of human capital in Ukraine. A comprehensive approach to human capital recovery is proposed, including investment in education and healthcare, support for migrants and internally displaced persons, the strengthening of civic responsibility and social cohesion, as well as the development and implementation of a social housing construction programme, which is considered essential for encouraging the return of Ukrainians from abroad. The role of education in the renewal and development of the country’s human capital is substantiated. Particular emphasis is placed on the need for synergy between educational, economic, and social public policies, the development of retraining programmes, the implementation of inclusive educational approaches, and the strengthening of cooperation between universities and the business sector. The article concludes that the recovery of human capital should become a key priority of state policy in the context of Ukraine’s post-war reconstruction. Such an approach will contribute not only to economic growth but also to strengthening societal resilience, preserving social cohesion, and ensuring the effective renewal and sustainable development of human capital.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The article examines the theoretical and practical foundations of human capital recovery as one of the key factors ensuring the economic and social development of Ukraine and highlights the role of education in this process. In the context of profound socio-economic transformations caused by the war, mass emigration, demographic losses, and the crisis of social institutions, the renewal and development of human capital have become issues of strategic importance. The author emphasizes that the loss of qualified professionals, forced migration, declining quality of education and healthcare, growing psychological exhaustion of the population, and the overall deterioration of the demographic situation significantly reduce the state’s capacity to respond effectively to emerging challenges and to sustain and renew human capital. The article argues that human capital should not be viewed solely as an economic category but rather as a multidimensional phenomenon that encompasses social, cultural, institutional, and value-based components. The relationship between the quality of human capital and the resilience of the entire social system at national, regional, and local community levels is analyzed. Given the contemporary challenges faced by Ukraine in wartime conditions, it is crucial to identify mechanisms that would enable the integration of human development policy into the national development strategy through a dedicated programme for the recovery and development of human capital in Ukraine. A comprehensive approach to human capital recovery is proposed, including investment in education and healthcare, support for migrants and internally displaced persons, the strengthening of civic responsibility and social cohesion, as well as the development and implementation of a social housing construction programme, which is considered essential for encouraging the return of Ukrainians from abroad. The role of education in the renewal and development of the country’s human capital is substantiated. Particular emphasis is placed on the need for synergy between educational, economic, and social public policies, the development of retraining programmes, the implementation of inclusive educational approaches, and the strengthening of cooperation between universities and the business sector. The article concludes that the recovery of human capital should become a key priority of state policy in the context of Ukraine’s post-war reconstruction. Such an approach will contribute not only to economic growth but also to strengthening societal resilience, preserving social cohesion, and ensuring the effective renewal and sustainable development of human capital.
13.
Xiong, Butian; Liu, Rong; Zhou, Tiantian; Chen, Meida; Fan, Zhiwen; Feng, Andrew
NanoGS: Training-Free Gaussian Splat Simplification Miscellaneous
2026, (arXiv:2603.16103 [cs.CV]).
@misc{xiong_nanogs_2026,
title = {NanoGS: Training-Free Gaussian Splat Simplification},
author = {Butian Xiong and Rong Liu and Tiantian Zhou and Meida Chen and Zhiwen Fan and Andrew Feng},
url = {http://arxiv.org/abs/2603.16103},
doi = {10.48550/arXiv.2603.16103},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-18},
publisher = {arXiv},
abstract = {3D Gaussian Splat (3DGS) enables high-fidelity, real-time novel view synthesis by representing scenes with large sets of anisotropic primitives, but often requires millions of Splats, incurring significant storage and transmission costs. Most existing compression methods rely on GPU-intensive post-training optimization with calibrated images, limiting practical deployment. We introduce NanoGS, a training-free and lightweight framework for Gaussian Splat simplification. Instead of relying on image-based rendering supervision, NanoGS formulates simplification as local pairwise merging over a sparse spatial graph. The method approximates a pair of Gaussians with a single primitive using mass preserved moment matching and evaluates merge quality through a principled merge cost between the original mixture and its approximation. By restricting merge candidates to local neighborhoods and selecting compatible pairs efficiently, NanoGS produces compact Gaussian representations while preserving scene structure and appearance. NanoGS operates directly on existing Gaussian Splat models, runs efficiently on CPU, and preserves the standard 3DGS parameterization, enabling seamless integration with existing rendering pipelines. Experiments demonstrate that NanoGS substantially reduces primitive count while maintaining high rendering fidelity, providing an efficient and practical solution for Gaussian Splat simplification. Our project website is available at https://saliteta.github.io/NanoGS/.},
note = {arXiv:2603.16103 [cs.CV]},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
3D Gaussian Splat (3DGS) enables high-fidelity, real-time novel view synthesis by representing scenes with large sets of anisotropic primitives, but often requires millions of Splats, incurring significant storage and transmission costs. Most existing compression methods rely on GPU-intensive post-training optimization with calibrated images, limiting practical deployment. We introduce NanoGS, a training-free and lightweight framework for Gaussian Splat simplification. Instead of relying on image-based rendering supervision, NanoGS formulates simplification as local pairwise merging over a sparse spatial graph. The method approximates a pair of Gaussians with a single primitive using mass preserved moment matching and evaluates merge quality through a principled merge cost between the original mixture and its approximation. By restricting merge candidates to local neighborhoods and selecting compatible pairs efficiently, NanoGS produces compact Gaussian representations while preserving scene structure and appearance. NanoGS operates directly on existing Gaussian Splat models, runs efficiently on CPU, and preserves the standard 3DGS parameterization, enabling seamless integration with existing rendering pipelines. Experiments demonstrate that NanoGS substantially reduces primitive count while maintaining high rendering fidelity, providing an efficient and practical solution for Gaussian Splat simplification. Our project website is available at https://saliteta.github.io/NanoGS/.
14.
Zhang, Tianyi; Traum, David
Rethinking Evaluation in Retrieval-Augmented Personalized Dialogue: A Cognitive and Linguistic Perspective Miscellaneous
2026, (arXiv:2603.14217 [cs.CL]).
@misc{zhang_rethinking_2026,
title = {Rethinking Evaluation in Retrieval-Augmented Personalized Dialogue: A Cognitive and Linguistic Perspective},
author = {Tianyi Zhang and David Traum},
url = {http://arxiv.org/abs/2603.14217},
doi = {10.48550/arXiv.2603.14217},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-18},
publisher = {arXiv},
abstract = {In cognitive science and linguistic theory, dialogue is not seen as a chain of independent utterances but rather as a joint activity sustained by coherence, consistency, and shared understanding. However, many systems for open-domain and personalized dialogue use surface-level similarity metrics (e.g., BLEU, ROUGE, F1) as one of their main reporting measures, which fail to capture these deeper aspects of conversational quality. We re-examine a notable retrieval-augmented framework for personalized dialogue, LAPDOG, as a case study for evaluation methodology. Using both human and LLM-based judges, we identify limitations in current evaluation practices, including corrupted dialogue histories, contradictions between retrieved stories and persona, and incoherent response generation. Our results show that human and LLM judgments align closely but diverge from lexical similarity metrics, underscoring the need for cognitively grounded evaluation methods. Broadly, this work charts a path toward more reliable assessment frameworks for retrieval-augmented dialogue systems that better reflect the principles of natural human communication.},
note = {arXiv:2603.14217 [cs.CL]},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
In cognitive science and linguistic theory, dialogue is not seen as a chain of independent utterances but rather as a joint activity sustained by coherence, consistency, and shared understanding. However, many systems for open-domain and personalized dialogue use surface-level similarity metrics (e.g., BLEU, ROUGE, F1) as one of their main reporting measures, which fail to capture these deeper aspects of conversational quality. We re-examine a notable retrieval-augmented framework for personalized dialogue, LAPDOG, as a case study for evaluation methodology. Using both human and LLM-based judges, we identify limitations in current evaluation practices, including corrupted dialogue histories, contradictions between retrieved stories and persona, and incoherent response generation. Our results show that human and LLM judgments align closely but diverge from lexical similarity metrics, underscoring the need for cognitively grounded evaluation methods. Broadly, this work charts a path toward more reliable assessment frameworks for retrieval-augmented dialogue systems that better reflect the principles of natural human communication.
15.
Hale, James; Dorley, Emmanuel J.; Kim, Peter H.; Gratch, Jonathan
IVAs to Study Gender Bias in Negotiation Journal Article
In: J Multimodal User Interfaces, vol. 20, no. 1, pp. 33–42, 2026, ISSN: 1783-7677, 1783-8738.
@article{hale_ivas_2026,
title = {IVAs to Study Gender Bias in Negotiation},
author = {James Hale and Emmanuel J. Dorley and Peter H. Kim and Jonathan Gratch},
url = {https://link.springer.com/10.1007/s12193-026-00476-5},
doi = {10.1007/s12193-026-00476-5},
issn = {1783-7677, 1783-8738},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-18},
journal = {J Multimodal User Interfaces},
volume = {20},
number = {1},
pages = {33–42},
abstract = {Abstract
Virtual agents offer a powerful and underutilized methodology for studying social interaction, particularly how bias and stereotypes operate. Drawing on theoretical frameworks such as the Media Equation and supported by empirical studies, we demonstrate how virtual agents can serve as controlled proxies for human social actors to study social bias. Through two empirical studies in the negotiation domain, we show how gender cues can be studied in negotiation contexts. Study 1 examines the effects of gendered job descriptions on negotiation behavior — finding that gendered language and gender interact to affect participants’ minimally acceptable salary. Study 2 explores how agent embodiment and emotional tone influence user perceptions across gender lines — finding that agent embodiment and one’s gender interact to affect participants’ subjective feelings about the dispute. Together, these findings imply the need for greater use of virtual agents in research, as they help inform the development of more inclusive and equitable AI systems.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Abstract
Virtual agents offer a powerful and underutilized methodology for studying social interaction, particularly how bias and stereotypes operate. Drawing on theoretical frameworks such as the Media Equation and supported by empirical studies, we demonstrate how virtual agents can serve as controlled proxies for human social actors to study social bias. Through two empirical studies in the negotiation domain, we show how gender cues can be studied in negotiation contexts. Study 1 examines the effects of gendered job descriptions on negotiation behavior — finding that gendered language and gender interact to affect participants’ minimally acceptable salary. Study 2 explores how agent embodiment and emotional tone influence user perceptions across gender lines — finding that agent embodiment and one’s gender interact to affect participants’ subjective feelings about the dispute. Together, these findings imply the need for greater use of virtual agents in research, as they help inform the development of more inclusive and equitable AI systems.
Virtual agents offer a powerful and underutilized methodology for studying social interaction, particularly how bias and stereotypes operate. Drawing on theoretical frameworks such as the Media Equation and supported by empirical studies, we demonstrate how virtual agents can serve as controlled proxies for human social actors to study social bias. Through two empirical studies in the negotiation domain, we show how gender cues can be studied in negotiation contexts. Study 1 examines the effects of gendered job descriptions on negotiation behavior — finding that gendered language and gender interact to affect participants’ minimally acceptable salary. Study 2 explores how agent embodiment and emotional tone influence user perceptions across gender lines — finding that agent embodiment and one’s gender interact to affect participants’ subjective feelings about the dispute. Together, these findings imply the need for greater use of virtual agents in research, as they help inform the development of more inclusive and equitable AI systems.
16.
Hu, Yue; Wang, Daniel; Song, Xiyun; Lao, Zhiqiang; Zhou, Yubin; Chen, Meida; Feng, Andrew; Yu, Heather
Diffuse4D:Completing NeRF-Stereo Depth via Diffusion-Driven Restoration in Dynamic Scenes Proceedings Article
In: pp. 1003–1011, 2026.
@inproceedings{hu_diffuse4dcompleting_2026,
title = {Diffuse4D:Completing NeRF-Stereo Depth via Diffusion-Driven Restoration in Dynamic Scenes},
author = {Yue Hu and Daniel Wang and Xiyun Song and Zhiqiang Lao and Yubin Zhou and Meida Chen and Andrew Feng and Heather Yu},
year = {2026},
date = {2026-03-01},
pages = {1003–1011},
abstract = {4D reconstruction recovers time-varying 3D geometry, yet
accurate depth estimation in moving regions remains challenging. Most existing methods—from monocular predictors to 4D Gaussian Splatting—rely on single-view priors prone to scale-shift ambiguity and geometric distortion.
While stereo estimation offers metric accuracy through triangulation, NeRF-based stereo synthesis often introduces
ghosting and motion blur in dynamic areas, degrading performance. To address this, we propose a hybrid framework
combining NeRF-generated stereo depth with a confidence-
aware diffusion module to restore dynamic geometry. Our
approach eliminates affine uncertainty and monocular dependence, achieving temporally coherent sequences. Experimental results show our method reduces absolute relative error by over 50% and improves strict accuracy (δ <
1.05) by 10%, setting a new benchmark for high-fidelity 4D
reconstruction.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
4D reconstruction recovers time-varying 3D geometry, yet
accurate depth estimation in moving regions remains challenging. Most existing methods—from monocular predictors to 4D Gaussian Splatting—rely on single-view priors prone to scale-shift ambiguity and geometric distortion.
While stereo estimation offers metric accuracy through triangulation, NeRF-based stereo synthesis often introduces
ghosting and motion blur in dynamic areas, degrading performance. To address this, we propose a hybrid framework
combining NeRF-generated stereo depth with a confidence-
aware diffusion module to restore dynamic geometry. Our
approach eliminates affine uncertainty and monocular dependence, achieving temporally coherent sequences. Experimental results show our method reduces absolute relative error by over 50% and improves strict accuracy (δ <
1.05) by 10%, setting a new benchmark for high-fidelity 4D
reconstruction.
accurate depth estimation in moving regions remains challenging. Most existing methods—from monocular predictors to 4D Gaussian Splatting—rely on single-view priors prone to scale-shift ambiguity and geometric distortion.
While stereo estimation offers metric accuracy through triangulation, NeRF-based stereo synthesis often introduces
ghosting and motion blur in dynamic areas, degrading performance. To address this, we propose a hybrid framework
combining NeRF-generated stereo depth with a confidence-
aware diffusion module to restore dynamic geometry. Our
approach eliminates affine uncertainty and monocular dependence, achieving temporally coherent sequences. Experimental results show our method reduces absolute relative error by over 50% and improves strict accuracy (δ <
1.05) by 10%, setting a new benchmark for high-fidelity 4D
reconstruction.
17.
O'Connell, Amy; Zhu, Xin; Kian, Mina; Dao, Warren; Gratch, Jonathan; Mataric, Maja; Culbertson, Heather
Exploring Remote Affective Communication Through a Haptic Wearable and Socially Assistive Robot Proceedings Article
In: Proceedings of the Twentieth International Conference on Tangible, Embedded, and Embodied Interaction, pp. 1–13, ACM, Chicago IL USA, 2026, ISBN: 979-8-4007-1868-7.
@inproceedings{oconnell_exploring_2026,
title = {Exploring Remote Affective Communication Through a Haptic Wearable and Socially Assistive Robot},
author = {Amy O'Connell and Xin Zhu and Mina Kian and Warren Dao and Jonathan Gratch and Maja Mataric and Heather Culbertson},
url = {https://dl.acm.org/doi/10.1145/3731459.3773338},
doi = {10.1145/3731459.3773338},
isbn = {979-8-4007-1868-7},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-19},
booktitle = {Proceedings of the Twentieth International Conference on Tangible, Embedded, and Embodied Interaction},
pages = {1–13},
publisher = {ACM},
address = {Chicago IL USA},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
18.
Tran, Minh; Siniukov, Maksim; Jin, Zhangyu; Soleymani, Mohammad
Discrete Facial Encoding: A Framework for the Data-driven Facial Display Discovery Proceedings Article
In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 2348–2358, 2026.
@inproceedings{tran_discrete_2026,
title = {Discrete Facial Encoding: A Framework for the Data-driven Facial Display Discovery},
author = {Minh Tran and Maksim Siniukov and Zhangyu Jin and Mohammad Soleymani},
year = {2026},
date = {2026-03-01},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
pages = {2348–2358},
abstract = {Facial expression analysis is central to understanding human behavior, yet existing coding systems such as the Facial Action Coding System (FACS) are constrained by limited coverage and costly manual annotation. In this work,
we introduce Discrete Facial Encoding (DFE), an unsupervised, data-driven alternative of compact and interpretable dictionary of facial expressions from 3D mesh sequences learned through a Residual Vector Quantized Variational Autoencoder (RVQ-VAE). Our approach first extracts identityinvariant expression features from images using a 3D Morphable Model (3DMM), effectively disentangling factors such as head pose and facial geometry. We then encode these features using an RVQ-VAE, producing a sequence of discrete tokens from a shared codebook, where each token captures a specific, reusable facial deformation patter that contributes to the overall expression. Through extensive experiments, we demonstrate that Discrete Facial Encoding captures more precise facial behaviors than FACS
and other facial encoding alternatives. We evaluate the
utility of our representation across three high-level psychological tasks: stress detection, personality prediction, and depression detection. Using a simple Bag-of-Words model built on top of the learned tokens, our system consistently outperforms both FACS-based pipelines and strong image and video representation learning models such as Masked Autoencoders. Further analysis reveals that our representation covers a wider variety of facial displays, highlighting its potential as a scalable and effective alternative to FACS for psychological and affective computing applications. We released the source code and model weights at https://github.com/ihp-lab/vqface.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Facial expression analysis is central to understanding human behavior, yet existing coding systems such as the Facial Action Coding System (FACS) are constrained by limited coverage and costly manual annotation. In this work,
we introduce Discrete Facial Encoding (DFE), an unsupervised, data-driven alternative of compact and interpretable dictionary of facial expressions from 3D mesh sequences learned through a Residual Vector Quantized Variational Autoencoder (RVQ-VAE). Our approach first extracts identityinvariant expression features from images using a 3D Morphable Model (3DMM), effectively disentangling factors such as head pose and facial geometry. We then encode these features using an RVQ-VAE, producing a sequence of discrete tokens from a shared codebook, where each token captures a specific, reusable facial deformation patter that contributes to the overall expression. Through extensive experiments, we demonstrate that Discrete Facial Encoding captures more precise facial behaviors than FACS
and other facial encoding alternatives. We evaluate the
utility of our representation across three high-level psychological tasks: stress detection, personality prediction, and depression detection. Using a simple Bag-of-Words model built on top of the learned tokens, our system consistently outperforms both FACS-based pipelines and strong image and video representation learning models such as Masked Autoencoders. Further analysis reveals that our representation covers a wider variety of facial displays, highlighting its potential as a scalable and effective alternative to FACS for psychological and affective computing applications. We released the source code and model weights at https://github.com/ihp-lab/vqface.
we introduce Discrete Facial Encoding (DFE), an unsupervised, data-driven alternative of compact and interpretable dictionary of facial expressions from 3D mesh sequences learned through a Residual Vector Quantized Variational Autoencoder (RVQ-VAE). Our approach first extracts identityinvariant expression features from images using a 3D Morphable Model (3DMM), effectively disentangling factors such as head pose and facial geometry. We then encode these features using an RVQ-VAE, producing a sequence of discrete tokens from a shared codebook, where each token captures a specific, reusable facial deformation patter that contributes to the overall expression. Through extensive experiments, we demonstrate that Discrete Facial Encoding captures more precise facial behaviors than FACS
and other facial encoding alternatives. We evaluate the
utility of our representation across three high-level psychological tasks: stress detection, personality prediction, and depression detection. Using a simple Bag-of-Words model built on top of the learned tokens, our system consistently outperforms both FACS-based pipelines and strong image and video representation learning models such as Masked Autoencoders. Further analysis reveals that our representation covers a wider variety of facial displays, highlighting its potential as a scalable and effective alternative to FACS for psychological and affective computing applications. We released the source code and model weights at https://github.com/ihp-lab/vqface.
19.
Xue, Jintang; Zhao, Ganning; Yao, Jie-En; Chen, Hong-En; Hu, Yue; Chen, Meida; You, Suya; Kuo, C. -C. Jay
Descrip3D: Enhancing Large Language Model-based 3D Scene Understanding with Object-Level Text Descriptions Proceedings Article
In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 1746–1756, 2026.
@inproceedings{xue_descrip3d_2026,
title = {Descrip3D: Enhancing Large Language Model-based 3D Scene Understanding with Object-Level Text Descriptions},
author = {Jintang Xue and Ganning Zhao and Jie-En Yao and Hong-En Chen and Yue Hu and Meida Chen and Suya You and C. -C. Jay Kuo},
year = {2026},
date = {2026-03-01},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
pages = {1746–1756},
abstract = {Understanding 3D scenes goes beyond simply recognizing objects; it requires reasoning about the spatial and semantic relationships between them. Current 3D scene-language models often struggle with this relational understanding, particularly when visual embeddings alone do not adequately convey the roles and interactions of objects. In this paper, we introduce Descrip3D, a novel and powerful framework that explicitly encodes the relationships between objects using natural language. Unlike previous methods that rely only on 2D and 3D embeddings, Descrip3D enhances each object with a textual description that captures both its intrinsic attributes and contextual relationships. These relational cues are incorporated into the model through a dual-level integration: embedding fusion and prompt-level injection. This allows for unified reasoning across various tasks such as grounding, captioning, and question answering, all without the need for task-specific heads or additional supervision. When evaluated on five benchmark datasets, including ScanRefer, Multi3DRefer, ScanQA, SQA3D, and Scan2Cap, Descrip3D consistently outperforms strong baseline models, demonstrating the effectiveness of language-guided relational representation for understanding complex indoor scenes.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Understanding 3D scenes goes beyond simply recognizing objects; it requires reasoning about the spatial and semantic relationships between them. Current 3D scene-language models often struggle with this relational understanding, particularly when visual embeddings alone do not adequately convey the roles and interactions of objects. In this paper, we introduce Descrip3D, a novel and powerful framework that explicitly encodes the relationships between objects using natural language. Unlike previous methods that rely only on 2D and 3D embeddings, Descrip3D enhances each object with a textual description that captures both its intrinsic attributes and contextual relationships. These relational cues are incorporated into the model through a dual-level integration: embedding fusion and prompt-level injection. This allows for unified reasoning across various tasks such as grounding, captioning, and question answering, all without the need for task-specific heads or additional supervision. When evaluated on five benchmark datasets, including ScanRefer, Multi3DRefer, ScanQA, SQA3D, and Scan2Cap, Descrip3D consistently outperforms strong baseline models, demonstrating the effectiveness of language-guided relational representation for understanding complex indoor scenes.
20.
Chaubey, Ashutosh; Guan, Xulang; Soleymani, Mohammad
Face-LLaVA: Facial Expression and Attribute Understanding through Instruction Tuning Proceedings Article
In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 2648–2660, 2026.
@inproceedings{chaubey_face-llava_2026,
title = {Face-LLaVA: Facial Expression and Attribute Understanding through Instruction Tuning},
author = {Ashutosh Chaubey and Xulang Guan and Mohammad Soleymani},
year = {2026},
date = {2026-03-01},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
pages = {2648–2660},
abstract = {The human face plays a central role in social communication, necessitating the use of performant computer vision tools for human-centered applications. We propose Face-LLaVA, a multimodal large language model for face-centered, in-context learning, including facial expression and attribute recognition. Additionally, Face-LLaVA is able to generate natural language descriptions that can be used for reasoning. Leveraging existing visual databases, we first developed FaceInstruct-1M, a face-centered database for instruction tuning MLLMs for face processing. We then developed a novel face-specific visual encoder powered by Face-Region Guided Cross-Attention that integrates face geometry with local visual features. We evaluated the proposed method across nine different datasets and five different face processing tasks, including facial expression recognition, action unit detection, facial attribute detection, age estimation and deepfake detection. Face-LLaVA achieves superior results compared to existing open-source MLLMs and competitive performance compared to commercial solutions. Our model output also receives a higher reasoning rating by GPT under a zero-shot setting across all the tasks. Both our dataset and model will be made public to support future advancements in social AI and foundational vision-language research.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The human face plays a central role in social communication, necessitating the use of performant computer vision tools for human-centered applications. We propose Face-LLaVA, a multimodal large language model for face-centered, in-context learning, including facial expression and attribute recognition. Additionally, Face-LLaVA is able to generate natural language descriptions that can be used for reasoning. Leveraging existing visual databases, we first developed FaceInstruct-1M, a face-centered database for instruction tuning MLLMs for face processing. We then developed a novel face-specific visual encoder powered by Face-Region Guided Cross-Attention that integrates face geometry with local visual features. We evaluated the proposed method across nine different datasets and five different face processing tasks, including facial expression recognition, action unit detection, facial attribute detection, age estimation and deepfake detection. Face-LLaVA achieves superior results compared to existing open-source MLLMs and competitive performance compared to commercial solutions. Our model output also receives a higher reasoning rating by GPT under a zero-shot setting across all the tasks. Both our dataset and model will be made public to support future advancements in social AI and foundational vision-language research.
Filter
2026
Barker, Sarah Alicia; Miles-Novelo, Andreas J; Rizzo, Albert; Tuma, Regina M
Efficacy of Virtual Reality–Based Mindfulness Interventions: Systematic Review and Meta-Analysis Journal Article
In: JMIR XR Spatial Comput, vol. 3, pp. e90003–e90003, 2026, ISSN: 2818-3045.
Abstract | Links | BibTeX | Tags: VR
@article{barker_efficacy_2026,
title = {Efficacy of Virtual Reality–Based Mindfulness Interventions: Systematic Review and Meta-Analysis},
author = {Sarah Alicia Barker and Andreas J Miles-Novelo and Albert Rizzo and Regina M Tuma},
url = {https://xr.jmir.org/2026/1/e90003},
doi = {10.2196/90003},
issn = {2818-3045},
year = {2026},
date = {2026-06-01},
urldate = {2026-07-09},
journal = {JMIR XR Spatial Comput},
volume = {3},
pages = {e90003–e90003},
abstract = {Background: Virtual reality–based mindfulness interventions (VRbMIs) increasingly populate studies as scalable tools for
stress and emotion regulation. However, findings across psychological outcomes are heterogeneous, and methodological
variation in intervention design, outcome measurement, and reporting practices limits cross-study comparability and cumulative synthesis.
Objective: A PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020–compliant systematic
review and meta-analysis was conducted to evaluate the psychological effects of VRbMIs published between 2023 and 2025,
examining methodological quality and outcome consistency across diverse study designs.
Methods: A PRISMA 2020–compliant systematic search was conducted across PubMed, PsycINFO, Scopus, Semantic
Scholar, and CORE from January 2023 to April 2025. Eligible studies included quantitative VRbMIs reporting psychological
outcomes. We assessed risk of bias using the Mixed Methods Appraisal Tool and conducted random-effects meta-analyses
using Hedges g where data were sufficient.},
keywords = {VR},
pubstate = {published},
tppubtype = {article}
}
Background: Virtual reality–based mindfulness interventions (VRbMIs) increasingly populate studies as scalable tools for
stress and emotion regulation. However, findings across psychological outcomes are heterogeneous, and methodological
variation in intervention design, outcome measurement, and reporting practices limits cross-study comparability and cumulative synthesis.
Objective: A PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020–compliant systematic
review and meta-analysis was conducted to evaluate the psychological effects of VRbMIs published between 2023 and 2025,
examining methodological quality and outcome consistency across diverse study designs.
Methods: A PRISMA 2020–compliant systematic search was conducted across PubMed, PsycINFO, Scopus, Semantic
Scholar, and CORE from January 2023 to April 2025. Eligible studies included quantitative VRbMIs reporting psychological
outcomes. We assessed risk of bias using the Mixed Methods Appraisal Tool and conducted random-effects meta-analyses
using Hedges g where data were sufficient.
stress and emotion regulation. However, findings across psychological outcomes are heterogeneous, and methodological
variation in intervention design, outcome measurement, and reporting practices limits cross-study comparability and cumulative synthesis.
Objective: A PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020–compliant systematic
review and meta-analysis was conducted to evaluate the psychological effects of VRbMIs published between 2023 and 2025,
examining methodological quality and outcome consistency across diverse study designs.
Methods: A PRISMA 2020–compliant systematic search was conducted across PubMed, PsycINFO, Scopus, Semantic
Scholar, and CORE from January 2023 to April 2025. Eligible studies included quantitative VRbMIs reporting psychological
outcomes. We assessed risk of bias using the Mixed Methods Appraisal Tool and conducted random-effects meta-analyses
using Hedges g where data were sufficient.
Zang, Tao; Awada, Mohamad; Becerik-Gerber, Burcin; Roll, Shawn C.; Lucas, Gale; Narayanan, Shrikanth
Does your office make you feel worse? Linking environmental exposure and temporal factors to cognitive symptoms Journal Article
In: Indoor Environments, vol. 3, no. 2, pp. 100171, 2026, ISSN: 29503620.
Abstract | Links | BibTeX | Tags: Machine Learning
@article{zang_does_2026,
title = {Does your office make you feel worse? Linking environmental exposure and temporal factors to cognitive symptoms},
author = {Tao Zang and Mohamad Awada and Burcin Becerik-Gerber and Shawn C. Roll and Gale Lucas and Shrikanth Narayanan},
url = {https://linkinghub.elsevier.com/retrieve/pii/S2950362026000238},
doi = {10.1016/j.indenv.2026.100171},
issn = {29503620},
year = {2026},
date = {2026-06-01},
urldate = {2026-06-18},
journal = {Indoor Environments},
volume = {3},
number = {2},
pages = {100171},
abstract = {Symptoms such as fatigue, headache, low motivation, and difficulty concentrating affect office workers’ cognitive and overall well-being. This study investigates the dynamic relationships among temporal patterns, indoor environmental quality (IEQ) factors, and cognitive symptoms in real offices. We collected about four months of continuous environmental data and ecological momentary assessments (EMAs) from ten office workers and integrated XGBoost and SHAP analysis to identify those relationships. Model performance varied by symptom and participant. Under random train-test splitting, accuracy generally exceeded 70% with F1 scores ranging from 40% to 80%. Performance declined under a temporal split, highlighting the real-world challenge of predicting future symptom states from past observations. Based on SHAP analysis, hour of day and day of week emerged as significant predictors, revealing temporal patterns in cognitive symptoms. Within observed one-hour windows prior to reported symptoms, higher relative humidity was generally associated with increased risk of cognitive symptoms, while higher illuminance was associated with reduced symptom risk. Associations with sound pressure level and TVOCs varied across participants, with no consistent directions. Our data-driven approach provides insights into temporal factors and environmental exposures associated with cognitive symptoms using naturalistic data. These findings highlight the combined influence of temporal and IEQ factors on cognitive symptoms and contribute to the development of adaptive, user-centric workplaces that support cognitive well-being management.},
keywords = {Machine Learning},
pubstate = {published},
tppubtype = {article}
}
Symptoms such as fatigue, headache, low motivation, and difficulty concentrating affect office workers’ cognitive and overall well-being. This study investigates the dynamic relationships among temporal patterns, indoor environmental quality (IEQ) factors, and cognitive symptoms in real offices. We collected about four months of continuous environmental data and ecological momentary assessments (EMAs) from ten office workers and integrated XGBoost and SHAP analysis to identify those relationships. Model performance varied by symptom and participant. Under random train-test splitting, accuracy generally exceeded 70% with F1 scores ranging from 40% to 80%. Performance declined under a temporal split, highlighting the real-world challenge of predicting future symptom states from past observations. Based on SHAP analysis, hour of day and day of week emerged as significant predictors, revealing temporal patterns in cognitive symptoms. Within observed one-hour windows prior to reported symptoms, higher relative humidity was generally associated with increased risk of cognitive symptoms, while higher illuminance was associated with reduced symptom risk. Associations with sound pressure level and TVOCs varied across participants, with no consistent directions. Our data-driven approach provides insights into temporal factors and environmental exposures associated with cognitive symptoms using naturalistic data. These findings highlight the combined influence of temporal and IEQ factors on cognitive symptoms and contribute to the development of adaptive, user-centric workplaces that support cognitive well-being management.
Cai, Yunxuan; Xiang, Sitao; Li, Zongjian; Chen, Haiwei; Zhao, Yajie
Bringing Diversity from Diffusion Models to Semantic-Guided Face Asset Generation Journal Article
In: ACM Trans. Graph., vol. 45, no. 3, pp. 1–19, 2026, ISSN: 0730-0301, 1557-7368.
Abstract | Links | BibTeX | Tags: ARL, DTIC, Graphics, UARC, VGL
@article{cai_bringing_2026,
title = {Bringing Diversity from Diffusion Models to Semantic-Guided Face Asset Generation},
author = {Yunxuan Cai and Sitao Xiang and Zongjian Li and Haiwei Chen and Yajie Zhao},
url = {https://dl.acm.org/doi/10.1145/3793859},
doi = {10.1145/3793859},
issn = {0730-0301, 1557-7368},
year = {2026},
date = {2026-06-01},
urldate = {2026-05-21},
journal = {ACM Trans. Graph.},
volume = {45},
number = {3},
pages = {1–19},
abstract = {High-quality 3D face asset creation remains costly due to reliance on controlled capture setups and manual processing, limiting scalability and diversity. We introduce a fully automated, semantically controllable framework for generating PBR-ready 3D facial assets without requiring dedicated scans. Our pipeline begins with a diffusion-based data synthesis stage, where 2D portrait samples from a pre-trained diffusion model are converted into 44K textured 3D face reconstructions via our proposed geometry recovery and texture normalization algorithm, which aligns arbitrarily shaded outputs into clean albedo space. Using this dataset, we train a disentangled adversarial generator that maps semantic attributes (age, gender, ethnicity) to UV-space geometry and albedo, enabling both direct sampling and continuous latent editing while preserving identity. A refinement stage further produces PBR materials and secondary assets (eyeballs, teeth, gums). The resulting system supports controllable face generation and post-editing in real time and exports directly to standard rendering and animation pipelines. We evaluate each component extensively and provide a web-based interactive interface to showcase practical deployment.},
keywords = {ARL, DTIC, Graphics, UARC, VGL},
pubstate = {published},
tppubtype = {article}
}
High-quality 3D face asset creation remains costly due to reliance on controlled capture setups and manual processing, limiting scalability and diversity. We introduce a fully automated, semantically controllable framework for generating PBR-ready 3D facial assets without requiring dedicated scans. Our pipeline begins with a diffusion-based data synthesis stage, where 2D portrait samples from a pre-trained diffusion model are converted into 44K textured 3D face reconstructions via our proposed geometry recovery and texture normalization algorithm, which aligns arbitrarily shaded outputs into clean albedo space. Using this dataset, we train a disentangled adversarial generator that maps semantic attributes (age, gender, ethnicity) to UV-space geometry and albedo, enabling both direct sampling and continuous latent editing while preserving identity. A refinement stage further produces PBR materials and secondary assets (eyeballs, teeth, gums). The resulting system supports controllable face generation and post-editing in real time and exports directly to standard rendering and animation pipelines. We evaluate each component extensively and provide a web-based interactive interface to showcase practical deployment.
Curhan, Jared R.; Gratch, Jonathan
Introduction to Special Issue Artificial Intelligence and Negotiation: A Framework for an Emerging Field Journal Article
In: Negotiation Journal, vol. 42, 2026, ISSN: 0748-4526, 1571-9979.
Abstract | Links | BibTeX | Tags: AI
@article{curhan_introduction_2026,
title = {Introduction to Special Issue Artificial Intelligence and Negotiation: A Framework for an Emerging Field},
author = {Jared R. Curhan and Jonathan Gratch},
url = {https://direct.mit.edu/ngtn/article/doi/10.1162/NGTN.a.69/136835/Introduction-to-Special-Issue-Artificial},
doi = {10.1162/NGTN.a.69},
issn = {0748-4526, 1571-9979},
year = {2026},
date = {2026-05-01},
urldate = {2026-06-16},
journal = {Negotiation Journal},
volume = {42},
abstract = {Abstract
Artificial intelligence is increasingly embedded in the practice, study, and teaching of negotiation, creating the need for a clear conceptual vocabulary for this emerging field. In this article, we propose a spatial and functional framework, identifying four primary roles AI may occupy in negotiation: as agent in front of the negotiator, as coach beside the negotiator, as backtable support behind the negotiator, and as mediator or facilitator between the parties. We also identify three broader domains in which AI is transforming the negotiation field: teaching, research, and real-world implementation. Although these categories are analytically distinct, they often overlap in practice, especially in field settings where multiple AI functions are integrated within the same system. By organizing these roles and domains within a single framework, we seek to clarify how AI is revolutionizing negotiation, and how its various applications relate to one another.
The significance of AI in negotiation lies not only in what these systems can do, but in how they are designed and positioned relative to human judgment, institutional context, and the conduct of negotiation itself. Drawing on the contributions in this special issue, we examine how AI is reshaping processes, expanding access to training, reorienting empirical research, and supporting practitioners in complex environments. At the same time, these developments raise enduring questions about delegation, transparency, equity, accountability, and the preservation of human judgment. We argue that the most promising trajectory for AI in negotiation is one of complementarity: systems designed to augment human capabilities, preserve meaningful human involvement, and strengthen negotiation practice rather than displace the human capacities that give it meaning. By clarifying roles and articulating normative principles for design and use, this framework offers both an analytic foundation for the field and guidance for its responsible development.},
keywords = {AI},
pubstate = {published},
tppubtype = {article}
}
Abstract
Artificial intelligence is increasingly embedded in the practice, study, and teaching of negotiation, creating the need for a clear conceptual vocabulary for this emerging field. In this article, we propose a spatial and functional framework, identifying four primary roles AI may occupy in negotiation: as agent in front of the negotiator, as coach beside the negotiator, as backtable support behind the negotiator, and as mediator or facilitator between the parties. We also identify three broader domains in which AI is transforming the negotiation field: teaching, research, and real-world implementation. Although these categories are analytically distinct, they often overlap in practice, especially in field settings where multiple AI functions are integrated within the same system. By organizing these roles and domains within a single framework, we seek to clarify how AI is revolutionizing negotiation, and how its various applications relate to one another.
The significance of AI in negotiation lies not only in what these systems can do, but in how they are designed and positioned relative to human judgment, institutional context, and the conduct of negotiation itself. Drawing on the contributions in this special issue, we examine how AI is reshaping processes, expanding access to training, reorienting empirical research, and supporting practitioners in complex environments. At the same time, these developments raise enduring questions about delegation, transparency, equity, accountability, and the preservation of human judgment. We argue that the most promising trajectory for AI in negotiation is one of complementarity: systems designed to augment human capabilities, preserve meaningful human involvement, and strengthen negotiation practice rather than displace the human capacities that give it meaning. By clarifying roles and articulating normative principles for design and use, this framework offers both an analytic foundation for the field and guidance for its responsible development.
Artificial intelligence is increasingly embedded in the practice, study, and teaching of negotiation, creating the need for a clear conceptual vocabulary for this emerging field. In this article, we propose a spatial and functional framework, identifying four primary roles AI may occupy in negotiation: as agent in front of the negotiator, as coach beside the negotiator, as backtable support behind the negotiator, and as mediator or facilitator between the parties. We also identify three broader domains in which AI is transforming the negotiation field: teaching, research, and real-world implementation. Although these categories are analytically distinct, they often overlap in practice, especially in field settings where multiple AI functions are integrated within the same system. By organizing these roles and domains within a single framework, we seek to clarify how AI is revolutionizing negotiation, and how its various applications relate to one another.
The significance of AI in negotiation lies not only in what these systems can do, but in how they are designed and positioned relative to human judgment, institutional context, and the conduct of negotiation itself. Drawing on the contributions in this special issue, we examine how AI is reshaping processes, expanding access to training, reorienting empirical research, and supporting practitioners in complex environments. At the same time, these developments raise enduring questions about delegation, transparency, equity, accountability, and the preservation of human judgment. We argue that the most promising trajectory for AI in negotiation is one of complementarity: systems designed to augment human capabilities, preserve meaningful human involvement, and strengthen negotiation practice rather than displace the human capacities that give it meaning. By clarifying roles and articulating normative principles for design and use, this framework offers both an analytic foundation for the field and guidance for its responsible development.
Han, Zicheng; Becerik-Gerber, Burcin; Liu, Ruying; Lucas, Gale; Doolab, Yogin; Zein, Sara; Atkins, Margo; Blass, Deborah; Miles, Lachlan
Emergency dynamics modeling with an agent-based framework: Active shooter scenario case study Journal Article
In: International Journal of Disaster Risk Reduction, vol. 142, pp. 106226, 2026, ISSN: 22124209.
Abstract | Links | BibTeX | Tags: Simulation, Virtual Humans
@article{han_emergency_2026,
title = {Emergency dynamics modeling with an agent-based framework: Active shooter scenario case study},
author = {Zicheng Han and Burcin Becerik-Gerber and Ruying Liu and Gale Lucas and Yogin Doolab and Sara Zein and Margo Atkins and Deborah Blass and Lachlan Miles},
url = {https://linkinghub.elsevier.com/retrieve/pii/S2212420926002384},
doi = {10.1016/j.ijdrr.2026.106226},
issn = {22124209},
year = {2026},
date = {2026-05-01},
urldate = {2026-06-16},
journal = {International Journal of Disaster Risk Reduction},
volume = {142},
pages = {106226},
abstract = {Modeling human-induced emergencies requires a structured representation of interactions among occupants, hazards, and built environments. While existing simulation approaches provide important insights, behavioral processes, hazard progression, and architectural conditions are often examined separately. This study develops a parameterized agent-based modeling framework that integrates heterogeneous occupant behaviors, hazard dynamics, and environmental configurations within a commercially available crowd-simulation platform to extend its usability within the academic and professional communities. The framework comprises environmental modeling, agent definition, interaction logic, risk evaluation, and simulation execution, enabling systematic configuration of behavioral rules and scenario parameters through customized scripting. An Active Shooter Incident (ASI) in a school building is simulated as a demonstrative case study to illustrate the framework's capability to configure and test emergency scenarios under different behavioral, hazard and environmental assumptions. Using the framework, the study examines the effects of classroom occupancy, door width, and door-locking mechanisms on occupant safety, evaluated using a token-based high-risk metric and factorial statistical analysis. Results indicate that higher occupancy increases exposure risk, timely door locking substantially improves safety outcomes, and the influence of door width is context-dependent. The consistency of these patterns with prior research supports the behavioral plausibility of the modeling approach. Overall, the study contributes a practical and extensible framework for operationalizing human–hazard–environment interactions in similar agent-based simulations and for generating scenario-specific insights to support design and preparedness strategies in human-induced emergency contexts.},
keywords = {Simulation, Virtual Humans},
pubstate = {published},
tppubtype = {article}
}
Modeling human-induced emergencies requires a structured representation of interactions among occupants, hazards, and built environments. While existing simulation approaches provide important insights, behavioral processes, hazard progression, and architectural conditions are often examined separately. This study develops a parameterized agent-based modeling framework that integrates heterogeneous occupant behaviors, hazard dynamics, and environmental configurations within a commercially available crowd-simulation platform to extend its usability within the academic and professional communities. The framework comprises environmental modeling, agent definition, interaction logic, risk evaluation, and simulation execution, enabling systematic configuration of behavioral rules and scenario parameters through customized scripting. An Active Shooter Incident (ASI) in a school building is simulated as a demonstrative case study to illustrate the framework's capability to configure and test emergency scenarios under different behavioral, hazard and environmental assumptions. Using the framework, the study examines the effects of classroom occupancy, door width, and door-locking mechanisms on occupant safety, evaluated using a token-based high-risk metric and factorial statistical analysis. Results indicate that higher occupancy increases exposure risk, timely door locking substantially improves safety outcomes, and the influence of door width is context-dependent. The consistency of these patterns with prior research supports the behavioral plausibility of the modeling approach. Overall, the study contributes a practical and extensible framework for operationalizing human–hazard–environment interactions in similar agent-based simulations and for generating scenario-specific insights to support design and preparedness strategies in human-induced emergency contexts.
Gratch, Jonathan
Agents at the Table: Balancing Strategy, Participation, and Cooperation in the Design of AI Negotiators and Mediators Journal Article
In: Negotiation Journal, vol. 42, 2026, ISSN: 0748-4526, 1571-9979.
Abstract | Links | BibTeX | Tags: AI, LLM
@article{gratch_agents_2026,
title = {Agents at the Table: Balancing Strategy, Participation, and Cooperation in the Design of AI Negotiators and Mediators},
author = {Jonathan Gratch},
url = {https://direct.mit.edu/ngtn/article/doi/10.1162/NGTN.a.65/136731/Agents-at-the-Table-Balancing-Strategy},
doi = {10.1162/NGTN.a.65},
issn = {0748-4526, 1571-9979},
year = {2026},
date = {2026-05-01},
urldate = {2026-06-16},
journal = {Negotiation Journal},
volume = {42},
abstract = {Abstract
Emerging advances in artificial intelligence—especially large language models—raise the prospect of AI systems acting as full participants in negotiation rather than merely supporting tools. This article surveys the evolution of automated negotiators and mediators and outlines three foundational design choices that shape their effectiveness, and indeed, how “effectiveness” is defined: whether systems should follow rational decision-making or mirror human relational and emotional behavior; whether people should delegate negotiation to autonomous agents or participate in the negotiation process; and how systems should balance self-interest with joint gains. Although these choices were once treated as dichotomous, recent systems increasingly integrate elements of each. These developments aim to align strategic decision-making with relational communication in ways that human counterparts judge as procedurally fair. The article illustrates these trends through four contributions presented at the 2025 AI Negotiation Summit at MIT.},
keywords = {AI, LLM},
pubstate = {published},
tppubtype = {article}
}
Abstract
Emerging advances in artificial intelligence—especially large language models—raise the prospect of AI systems acting as full participants in negotiation rather than merely supporting tools. This article surveys the evolution of automated negotiators and mediators and outlines three foundational design choices that shape their effectiveness, and indeed, how “effectiveness” is defined: whether systems should follow rational decision-making or mirror human relational and emotional behavior; whether people should delegate negotiation to autonomous agents or participate in the negotiation process; and how systems should balance self-interest with joint gains. Although these choices were once treated as dichotomous, recent systems increasingly integrate elements of each. These developments aim to align strategic decision-making with relational communication in ways that human counterparts judge as procedurally fair. The article illustrates these trends through four contributions presented at the 2025 AI Negotiation Summit at MIT.
Emerging advances in artificial intelligence—especially large language models—raise the prospect of AI systems acting as full participants in negotiation rather than merely supporting tools. This article surveys the evolution of automated negotiators and mediators and outlines three foundational design choices that shape their effectiveness, and indeed, how “effectiveness” is defined: whether systems should follow rational decision-making or mirror human relational and emotional behavior; whether people should delegate negotiation to autonomous agents or participate in the negotiation process; and how systems should balance self-interest with joint gains. Although these choices were once treated as dichotomous, recent systems increasingly integrate elements of each. These developments aim to align strategic decision-making with relational communication in ways that human counterparts judge as procedurally fair. The article illustrates these trends through four contributions presented at the 2025 AI Negotiation Summit at MIT.
Hans, Soham; Wang, Yunzhe; Ustun, Volkan
Egocentric Team AI: Enabling Tactical Reasoning from the Operator’s View Journal Article
In: AAAI-SS, vol. 8, no. 1, pp. 45–49, 2026, ISSN: 2994-4317.
Abstract | Links | BibTeX | Tags: CogArch, DTIC, UARC
@article{hans_egocentric_2026,
title = {Egocentric Team AI: Enabling Tactical Reasoning from the Operator’s View},
author = {Soham Hans and Yunzhe Wang and Volkan Ustun},
url = {https://ojs.aaai.org/index.php/AAAI-SS/article/view/42516},
doi = {10.1609/aaaiss.v8i1.42516},
issn = {2994-4317},
year = {2026},
date = {2026-05-01},
urldate = {2026-06-16},
journal = {AAAI-SS},
volume = {8},
number = {1},
pages = {45–49},
abstract = {The future of human–AI integration in high-stakes environments—ranging from defense operations to emergency response—requires AI systems that function as intuitive teammates rather than isolated tools. Yet a persistent perspective gap remains: video understanding models rely on third-person, broadcast-style views, while multi-agent reinforcement learning (MARL) systems operate on egocentric inputs but often depend on centralized critics during training, reducing the need for decentralized policies to internalize team-centric structure. As a result, agents may optimize behavior without learning policy-level representations of the tactical picture grounded in their own observations. To address this gap, we propose Egocentric Team AI, a research direction centered on learning implicit, distributed Common Operating Pictures from first-person views. Building on our prior work with the X-Ego-CS dataset and Cross-Ego Contrastive Learning (CECL), we outline how cross-egocentric representation alignment can be extended from passive video understanding to active multi-agent control. Specifically, we propose integrating cross-ego contrastive objectives into MARL within a multi-agent Doom-based environment as a testbed for decentralized, team-aware policy learning. By positioning cross-egocentric alignment as an inductive bias for decentralized coordination, this work charts a path toward embodied systems capable of adaptive cooperation without explicit communication.},
keywords = {CogArch, DTIC, UARC},
pubstate = {published},
tppubtype = {article}
}
The future of human–AI integration in high-stakes environments—ranging from defense operations to emergency response—requires AI systems that function as intuitive teammates rather than isolated tools. Yet a persistent perspective gap remains: video understanding models rely on third-person, broadcast-style views, while multi-agent reinforcement learning (MARL) systems operate on egocentric inputs but often depend on centralized critics during training, reducing the need for decentralized policies to internalize team-centric structure. As a result, agents may optimize behavior without learning policy-level representations of the tactical picture grounded in their own observations. To address this gap, we propose Egocentric Team AI, a research direction centered on learning implicit, distributed Common Operating Pictures from first-person views. Building on our prior work with the X-Ego-CS dataset and Cross-Ego Contrastive Learning (CECL), we outline how cross-egocentric representation alignment can be extended from passive video understanding to active multi-agent control. Specifically, we propose integrating cross-ego contrastive objectives into MARL within a multi-agent Doom-based environment as a testbed for decentralized, team-aware policy learning. By positioning cross-egocentric alignment as an inductive bias for decentralized coordination, this work charts a path toward embodied systems capable of adaptive cooperation without explicit communication.
Mozgai, Sharon; Fast, Ed; Leeds, Andrew; Sookiassian, Edwin; Kim, Kevin; Hartholt, Arno
Rapid personalized doppelgänger avatar generation: Dyadic evaluation of the TAC-Twin virtual human pipeline. Proceedings Article
In: 2026.
Abstract | Links | BibTeX | Tags: Virtual Humans
@inproceedings{mozgai_rapid_2026,
title = {Rapid personalized doppelgänger avatar generation: Dyadic evaluation of the TAC-Twin virtual human pipeline.},
author = {Sharon Mozgai and Ed Fast and Andrew Leeds and Edwin Sookiassian and Kevin Kim and Arno Hartholt},
url = {https://openaccess.cms-conferences.org/publications/book/978-1-964867-77-9/article/978-1-964867-77-9_39},
doi = {10.54941/ahfe1007194},
year = {2026},
date = {2026-04-01},
urldate = {2026-06-18},
abstract = {Rapid generation of personalized doppelgänger avatars remains challenging due to the
time, expertise, and system integration typically required to produce virtual humans
suitable for interactive use. TAC-Twin is a modular avatar generation pipeline developed
within the Virtual Human Toolkit (VHToolkit) that converts a single photograph into a
fully rigged avatar deployable in Unity-based virtual human systems using productionready tools and an interoperable architecture. This paper presents an exploratory dyadic
evaluation of TAC-Twin to examine how the pipeline performs in practice when generating
self-, partner-, and generic avatars for a standardized scenario. Findings indicate
that the pipeline supports efficient, repeatable avatar creation and produces visually
recognizable avatars appropriate for research deployment. Participant feedback further
suggests that perceived avatar quality was shaped not only by facial resemblance, but
by how personalization carried through to facial animation, movement, and behavioral
coherence during scenario playback. The evaluation highlights that advances in avatar
generation alone are insufficient to sustain perceived realism over repeated exposure
if downstream expressive and behavioral layers are not aligned, underscoring the
importance of end-to-end pipeline design for personalized virtual humans.},
keywords = {Virtual Humans},
pubstate = {published},
tppubtype = {inproceedings}
}
Rapid generation of personalized doppelgänger avatars remains challenging due to the
time, expertise, and system integration typically required to produce virtual humans
suitable for interactive use. TAC-Twin is a modular avatar generation pipeline developed
within the Virtual Human Toolkit (VHToolkit) that converts a single photograph into a
fully rigged avatar deployable in Unity-based virtual human systems using productionready tools and an interoperable architecture. This paper presents an exploratory dyadic
evaluation of TAC-Twin to examine how the pipeline performs in practice when generating
self-, partner-, and generic avatars for a standardized scenario. Findings indicate
that the pipeline supports efficient, repeatable avatar creation and produces visually
recognizable avatars appropriate for research deployment. Participant feedback further
suggests that perceived avatar quality was shaped not only by facial resemblance, but
by how personalization carried through to facial animation, movement, and behavioral
coherence during scenario playback. The evaluation highlights that advances in avatar
generation alone are insufficient to sustain perceived realism over repeated exposure
if downstream expressive and behavioral layers are not aligned, underscoring the
importance of end-to-end pipeline design for personalized virtual humans.
time, expertise, and system integration typically required to produce virtual humans
suitable for interactive use. TAC-Twin is a modular avatar generation pipeline developed
within the Virtual Human Toolkit (VHToolkit) that converts a single photograph into a
fully rigged avatar deployable in Unity-based virtual human systems using productionready tools and an interoperable architecture. This paper presents an exploratory dyadic
evaluation of TAC-Twin to examine how the pipeline performs in practice when generating
self-, partner-, and generic avatars for a standardized scenario. Findings indicate
that the pipeline supports efficient, repeatable avatar creation and produces visually
recognizable avatars appropriate for research deployment. Participant feedback further
suggests that perceived avatar quality was shaped not only by facial resemblance, but
by how personalization carried through to facial animation, movement, and behavioral
coherence during scenario playback. The evaluation highlights that advances in avatar
generation alone are insufficient to sustain perceived realism over repeated exposure
if downstream expressive and behavioral layers are not aligned, underscoring the
importance of end-to-end pipeline design for personalized virtual humans.
Wang, Yunzhe; Xu, Runhui; Zheng, Kexin; Zhang, Tianyi; Kogundi, Jayavibhav Niranjan; Hans, Soham; Ustun, Volkan
GameplayQA: A Benchmarking Framework for Decision-Dense POV-Synced Multi-Video Understanding of 3D Virtual Agents Miscellaneous
2026, (arXiv:2603.24329 [cs.CL]).
Abstract | Links | BibTeX | Tags: AI, DTIC, LLM, UARC, Virtual Agents
@misc{wang_gameplayqa_2026,
title = {GameplayQA: A Benchmarking Framework for Decision-Dense POV-Synced Multi-Video Understanding of 3D Virtual Agents},
author = {Yunzhe Wang and Runhui Xu and Kexin Zheng and Tianyi Zhang and Jayavibhav Niranjan Kogundi and Soham Hans and Volkan Ustun},
url = {http://arxiv.org/abs/2603.24329},
doi = {10.48550/arXiv.2603.24329},
year = {2026},
date = {2026-04-01},
urldate = {2026-06-18},
publisher = {arXiv},
abstract = {Multimodal LLMs are increasingly deployed as perceptual backbones for autonomous agents in 3D environments, from robotics to virtual worlds. These applications require agents to perceive rapid state changes, attribute actions to the correct entities, and reason about concurrent multi-agent behaviors from a first-person perspective, capabilities that existing benchmarks do not adequately evaluate. We introduce GameplayQA, a framework for evaluating agentic-centric perception and reasoning through video understanding. Specifically, we densely annotate multiplayer 3D gameplay videos at 1.22 labels/second, with time-synced, concurrent captions of states, actions, and events structured around a triadic system of Self, Other Agents, and the World, a natural decomposition for multi-agent environments. From these annotations, we refined 2.4K diagnostic QA pairs organized into three levels of cognitive complexity, accompanied by a structured distractor taxonomy that enables fine-grained analysis of where models hallucinate. Evaluation of frontier MLLMs reveals a substantial gap from human performance, with common failures in temporal and cross-video grounding, agent-role attribution, and handling the decision density of the game. We hope GameplayQA stimulates future research at the intersection of embodied AI, agentic perception, and world modeling.},
note = {arXiv:2603.24329 [cs.CL]},
keywords = {AI, DTIC, LLM, UARC, Virtual Agents},
pubstate = {published},
tppubtype = {misc}
}
Multimodal LLMs are increasingly deployed as perceptual backbones for autonomous agents in 3D environments, from robotics to virtual worlds. These applications require agents to perceive rapid state changes, attribute actions to the correct entities, and reason about concurrent multi-agent behaviors from a first-person perspective, capabilities that existing benchmarks do not adequately evaluate. We introduce GameplayQA, a framework for evaluating agentic-centric perception and reasoning through video understanding. Specifically, we densely annotate multiplayer 3D gameplay videos at 1.22 labels/second, with time-synced, concurrent captions of states, actions, and events structured around a triadic system of Self, Other Agents, and the World, a natural decomposition for multi-agent environments. From these annotations, we refined 2.4K diagnostic QA pairs organized into three levels of cognitive complexity, accompanied by a structured distractor taxonomy that enables fine-grained analysis of where models hallucinate. Evaluation of frontier MLLMs reveals a substantial gap from human performance, with common failures in temporal and cross-video grounding, agent-role attribution, and handling the decision density of the game. We hope GameplayQA stimulates future research at the intersection of embodied AI, agentic perception, and world modeling.
Xiong, Xinmiao; Liu, Bangya; Wang, Hao; Li, Dayou; Chen, Nuo; Feng, Andrew; Ding, Mingyu; Banerjee, Suman; Zhou, Yang; Fan, Zhiwen
Accelerating Transformer-Based Monocular SLAM via Geometric Utility Scoring Miscellaneous
2026, (arXiv:2604.08718 [cs.CV]).
Abstract | Links | BibTeX | Tags: AR, Geospatial
@misc{xiong_accelerating_2026,
title = {Accelerating Transformer-Based Monocular SLAM via Geometric Utility Scoring},
author = {Xinmiao Xiong and Bangya Liu and Hao Wang and Dayou Li and Nuo Chen and Andrew Feng and Mingyu Ding and Suman Banerjee and Yang Zhou and Zhiwen Fan},
url = {http://arxiv.org/abs/2604.08718},
doi = {10.48550/arXiv.2604.08718},
year = {2026},
date = {2026-04-01},
urldate = {2026-06-18},
publisher = {arXiv},
abstract = {Geometric Foundation Models (GFMs) have recently advanced monocular SLAM by providing robust, calibration-free 3D priors. However, deploying these models on dense video streams introduces significant computational redundancy. Current GFM-based SLAM systems typically rely on post hoc keyframe selection. Because of this, they must perform expensive dense geometric decoding simply to determine whether a frame contains novel geometry, resulting in late rejection and wasted computation. To mitigate this inefficiency, we propose LeanGate, a lightweight feed-forward frame-gating network. LeanGate predicts a geometric utility score to assess a frame's mapping value prior to the heavy GFM feature extraction and matching stages. As a predictive plug-and-play module, our approach bypasses over 90% of redundant frames. Evaluations on standard SLAM benchmarks demonstrate that LeanGate reduces tracking FLOPs by more than 85% and achieves a 5x end-to-end throughput speedup. Furthermore, it maintains the tracking and mapping accuracy of dense baselines. Project page: https://lean-gate.github.io/},
note = {arXiv:2604.08718 [cs.CV]},
keywords = {AR, Geospatial},
pubstate = {published},
tppubtype = {misc}
}
Geometric Foundation Models (GFMs) have recently advanced monocular SLAM by providing robust, calibration-free 3D priors. However, deploying these models on dense video streams introduces significant computational redundancy. Current GFM-based SLAM systems typically rely on post hoc keyframe selection. Because of this, they must perform expensive dense geometric decoding simply to determine whether a frame contains novel geometry, resulting in late rejection and wasted computation. To mitigate this inefficiency, we propose LeanGate, a lightweight feed-forward frame-gating network. LeanGate predicts a geometric utility score to assess a frame's mapping value prior to the heavy GFM feature extraction and matching stages. As a predictive plug-and-play module, our approach bypasses over 90% of redundant frames. Evaluations on standard SLAM benchmarks demonstrate that LeanGate reduces tracking FLOPs by more than 85% and achieves a 5x end-to-end throughput speedup. Furthermore, it maintains the tracking and mapping accuracy of dense baselines. Project page: https://lean-gate.github.io/
Feng, Andrew Zhuoer; Wang, Cunxiang; Luo, Yu; Fan, Lin; Zhou, Yilin; Wang, Zikang; Gu, Xiaotao; Tang, Jie; Wang, Hongning; Huang, Minlie
HoWToBench: Holistic Evaluation for LLM's Capability in Human-level Writing using Tree of Writing Miscellaneous
2026, (arXiv:2604.19071 [cs.CL]).
Abstract | Links | BibTeX | Tags: LLM
@misc{feng_howtobench_2026,
title = {HoWToBench: Holistic Evaluation for LLM's Capability in Human-level Writing using Tree of Writing},
author = {Andrew Zhuoer Feng and Cunxiang Wang and Yu Luo and Lin Fan and Yilin Zhou and Zikang Wang and Xiaotao Gu and Jie Tang and Hongning Wang and Minlie Huang},
url = {http://arxiv.org/abs/2604.19071},
doi = {10.48550/arXiv.2604.19071},
year = {2026},
date = {2026-04-01},
urldate = {2026-06-17},
publisher = {arXiv},
abstract = {Evaluating the writing capabilities of large language models (LLMs) remains a significant challenge due to the multidimensional nature of writing skills and the limitations of existing metrics. LLM's performance in thousand-words level and open-ended writing is inadequately assessed by traditional reference-based metrics or modern LLM-as-a-judge methods. We propose Tree-of-Writing (ToW), to resolve the implicit inconsistency often found when LLM-as-a-judge aggregates all sub-features in text evaluation. ToW incorporates a tree-structured workflow by explicitly modeling the aggregation weights of sub-features. We also present HowToBench, a large-scale Chinese writing benchmark encompassing 12 genres and 1302 instructions across three task categories: contextual completion, outline-guided writing, and open-ended generation. ToW successfully mitigates the biases, achieving a 0.93 Pearson correlation with human judgments. Furthermore, we detect that both overlap-based text generation metrics and popular LLM-as-a-judge practices are vulnerable to textual disturbances, while ToW is robust to them. We also uncover a negative correlation between input length and content-related scores in the Guide task, showcasing that it cannot be simply improved by input-side information piling.},
note = {arXiv:2604.19071 [cs.CL]},
keywords = {LLM},
pubstate = {published},
tppubtype = {misc}
}
Evaluating the writing capabilities of large language models (LLMs) remains a significant challenge due to the multidimensional nature of writing skills and the limitations of existing metrics. LLM's performance in thousand-words level and open-ended writing is inadequately assessed by traditional reference-based metrics or modern LLM-as-a-judge methods. We propose Tree-of-Writing (ToW), to resolve the implicit inconsistency often found when LLM-as-a-judge aggregates all sub-features in text evaluation. ToW incorporates a tree-structured workflow by explicitly modeling the aggregation weights of sub-features. We also present HowToBench, a large-scale Chinese writing benchmark encompassing 12 genres and 1302 instructions across three task categories: contextual completion, outline-guided writing, and open-ended generation. ToW successfully mitigates the biases, achieving a 0.93 Pearson correlation with human judgments. Furthermore, we detect that both overlap-based text generation metrics and popular LLM-as-a-judge practices are vulnerable to textual disturbances, while ToW is robust to them. We also uncover a negative correlation between input length and content-related scores in the Guide task, showcasing that it cannot be simply improved by input-side information piling.
Melnychenko, Olga
HUMAN CAPITAL IN UKRAINE: PROBLEMS OF RECOVERY AND DEVELOPMENT (THE EDUCOLOGICAL ASPECTS) Journal Article
In: Continuing Professional Education Theory and Practice, pp. 7, 2026, ISSN: 1609-8595, 2412-0774.
Abstract | Links | BibTeX | Tags: MedVR, VR
@article{melnychenko_human_2026,
title = {HUMAN CAPITAL IN UKRAINE: PROBLEMS OF RECOVERY AND DEVELOPMENT (THE EDUCOLOGICAL ASPECTS)},
author = {Olga Melnychenko},
url = {https://npo.kubg.edu.ua/article/view/354893},
doi = {10.28925/2412-0774.2026.1.1},
issn = {1609-8595, 2412-0774},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-18},
journal = {Continuing Professional Education Theory and Practice},
pages = {7},
abstract = {The article examines the theoretical and practical foundations of human capital recovery as one of the key factors ensuring the economic and social development of Ukraine and highlights the role of education in this process. In the context of profound socio-economic transformations caused by the war, mass emigration, demographic losses, and the crisis of social institutions, the renewal and development of human capital have become issues of strategic importance. The author emphasizes that the loss of qualified professionals, forced migration, declining quality of education and healthcare, growing psychological exhaustion of the population, and the overall deterioration of the demographic situation significantly reduce the state’s capacity to respond effectively to emerging challenges and to sustain and renew human capital. The article argues that human capital should not be viewed solely as an economic category but rather as a multidimensional phenomenon that encompasses social, cultural, institutional, and value-based components. The relationship between the quality of human capital and the resilience of the entire social system at national, regional, and local community levels is analyzed. Given the contemporary challenges faced by Ukraine in wartime conditions, it is crucial to identify mechanisms that would enable the integration of human development policy into the national development strategy through a dedicated programme for the recovery and development of human capital in Ukraine. A comprehensive approach to human capital recovery is proposed, including investment in education and healthcare, support for migrants and internally displaced persons, the strengthening of civic responsibility and social cohesion, as well as the development and implementation of a social housing construction programme, which is considered essential for encouraging the return of Ukrainians from abroad. The role of education in the renewal and development of the country’s human capital is substantiated. Particular emphasis is placed on the need for synergy between educational, economic, and social public policies, the development of retraining programmes, the implementation of inclusive educational approaches, and the strengthening of cooperation between universities and the business sector. The article concludes that the recovery of human capital should become a key priority of state policy in the context of Ukraine’s post-war reconstruction. Such an approach will contribute not only to economic growth but also to strengthening societal resilience, preserving social cohesion, and ensuring the effective renewal and sustainable development of human capital.},
keywords = {MedVR, VR},
pubstate = {published},
tppubtype = {article}
}
The article examines the theoretical and practical foundations of human capital recovery as one of the key factors ensuring the economic and social development of Ukraine and highlights the role of education in this process. In the context of profound socio-economic transformations caused by the war, mass emigration, demographic losses, and the crisis of social institutions, the renewal and development of human capital have become issues of strategic importance. The author emphasizes that the loss of qualified professionals, forced migration, declining quality of education and healthcare, growing psychological exhaustion of the population, and the overall deterioration of the demographic situation significantly reduce the state’s capacity to respond effectively to emerging challenges and to sustain and renew human capital. The article argues that human capital should not be viewed solely as an economic category but rather as a multidimensional phenomenon that encompasses social, cultural, institutional, and value-based components. The relationship between the quality of human capital and the resilience of the entire social system at national, regional, and local community levels is analyzed. Given the contemporary challenges faced by Ukraine in wartime conditions, it is crucial to identify mechanisms that would enable the integration of human development policy into the national development strategy through a dedicated programme for the recovery and development of human capital in Ukraine. A comprehensive approach to human capital recovery is proposed, including investment in education and healthcare, support for migrants and internally displaced persons, the strengthening of civic responsibility and social cohesion, as well as the development and implementation of a social housing construction programme, which is considered essential for encouraging the return of Ukrainians from abroad. The role of education in the renewal and development of the country’s human capital is substantiated. Particular emphasis is placed on the need for synergy between educational, economic, and social public policies, the development of retraining programmes, the implementation of inclusive educational approaches, and the strengthening of cooperation between universities and the business sector. The article concludes that the recovery of human capital should become a key priority of state policy in the context of Ukraine’s post-war reconstruction. Such an approach will contribute not only to economic growth but also to strengthening societal resilience, preserving social cohesion, and ensuring the effective renewal and sustainable development of human capital.
Xiong, Butian; Liu, Rong; Zhou, Tiantian; Chen, Meida; Fan, Zhiwen; Feng, Andrew
NanoGS: Training-Free Gaussian Splat Simplification Miscellaneous
2026, (arXiv:2603.16103 [cs.CV]).
Abstract | Links | BibTeX | Tags: Geospatial, Graphics
@misc{xiong_nanogs_2026,
title = {NanoGS: Training-Free Gaussian Splat Simplification},
author = {Butian Xiong and Rong Liu and Tiantian Zhou and Meida Chen and Zhiwen Fan and Andrew Feng},
url = {http://arxiv.org/abs/2603.16103},
doi = {10.48550/arXiv.2603.16103},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-18},
publisher = {arXiv},
abstract = {3D Gaussian Splat (3DGS) enables high-fidelity, real-time novel view synthesis by representing scenes with large sets of anisotropic primitives, but often requires millions of Splats, incurring significant storage and transmission costs. Most existing compression methods rely on GPU-intensive post-training optimization with calibrated images, limiting practical deployment. We introduce NanoGS, a training-free and lightweight framework for Gaussian Splat simplification. Instead of relying on image-based rendering supervision, NanoGS formulates simplification as local pairwise merging over a sparse spatial graph. The method approximates a pair of Gaussians with a single primitive using mass preserved moment matching and evaluates merge quality through a principled merge cost between the original mixture and its approximation. By restricting merge candidates to local neighborhoods and selecting compatible pairs efficiently, NanoGS produces compact Gaussian representations while preserving scene structure and appearance. NanoGS operates directly on existing Gaussian Splat models, runs efficiently on CPU, and preserves the standard 3DGS parameterization, enabling seamless integration with existing rendering pipelines. Experiments demonstrate that NanoGS substantially reduces primitive count while maintaining high rendering fidelity, providing an efficient and practical solution for Gaussian Splat simplification. Our project website is available at https://saliteta.github.io/NanoGS/.},
note = {arXiv:2603.16103 [cs.CV]},
keywords = {Geospatial, Graphics},
pubstate = {published},
tppubtype = {misc}
}
3D Gaussian Splat (3DGS) enables high-fidelity, real-time novel view synthesis by representing scenes with large sets of anisotropic primitives, but often requires millions of Splats, incurring significant storage and transmission costs. Most existing compression methods rely on GPU-intensive post-training optimization with calibrated images, limiting practical deployment. We introduce NanoGS, a training-free and lightweight framework for Gaussian Splat simplification. Instead of relying on image-based rendering supervision, NanoGS formulates simplification as local pairwise merging over a sparse spatial graph. The method approximates a pair of Gaussians with a single primitive using mass preserved moment matching and evaluates merge quality through a principled merge cost between the original mixture and its approximation. By restricting merge candidates to local neighborhoods and selecting compatible pairs efficiently, NanoGS produces compact Gaussian representations while preserving scene structure and appearance. NanoGS operates directly on existing Gaussian Splat models, runs efficiently on CPU, and preserves the standard 3DGS parameterization, enabling seamless integration with existing rendering pipelines. Experiments demonstrate that NanoGS substantially reduces primitive count while maintaining high rendering fidelity, providing an efficient and practical solution for Gaussian Splat simplification. Our project website is available at https://saliteta.github.io/NanoGS/.
Zhang, Tianyi; Traum, David
Rethinking Evaluation in Retrieval-Augmented Personalized Dialogue: A Cognitive and Linguistic Perspective Miscellaneous
2026, (arXiv:2603.14217 [cs.CL]).
Abstract | Links | BibTeX | Tags: ARO, Dialogue, DTIC, Natural Language, UARC
@misc{zhang_rethinking_2026,
title = {Rethinking Evaluation in Retrieval-Augmented Personalized Dialogue: A Cognitive and Linguistic Perspective},
author = {Tianyi Zhang and David Traum},
url = {http://arxiv.org/abs/2603.14217},
doi = {10.48550/arXiv.2603.14217},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-18},
publisher = {arXiv},
abstract = {In cognitive science and linguistic theory, dialogue is not seen as a chain of independent utterances but rather as a joint activity sustained by coherence, consistency, and shared understanding. However, many systems for open-domain and personalized dialogue use surface-level similarity metrics (e.g., BLEU, ROUGE, F1) as one of their main reporting measures, which fail to capture these deeper aspects of conversational quality. We re-examine a notable retrieval-augmented framework for personalized dialogue, LAPDOG, as a case study for evaluation methodology. Using both human and LLM-based judges, we identify limitations in current evaluation practices, including corrupted dialogue histories, contradictions between retrieved stories and persona, and incoherent response generation. Our results show that human and LLM judgments align closely but diverge from lexical similarity metrics, underscoring the need for cognitively grounded evaluation methods. Broadly, this work charts a path toward more reliable assessment frameworks for retrieval-augmented dialogue systems that better reflect the principles of natural human communication.},
note = {arXiv:2603.14217 [cs.CL]},
keywords = {ARO, Dialogue, DTIC, Natural Language, UARC},
pubstate = {published},
tppubtype = {misc}
}
In cognitive science and linguistic theory, dialogue is not seen as a chain of independent utterances but rather as a joint activity sustained by coherence, consistency, and shared understanding. However, many systems for open-domain and personalized dialogue use surface-level similarity metrics (e.g., BLEU, ROUGE, F1) as one of their main reporting measures, which fail to capture these deeper aspects of conversational quality. We re-examine a notable retrieval-augmented framework for personalized dialogue, LAPDOG, as a case study for evaluation methodology. Using both human and LLM-based judges, we identify limitations in current evaluation practices, including corrupted dialogue histories, contradictions between retrieved stories and persona, and incoherent response generation. Our results show that human and LLM judgments align closely but diverge from lexical similarity metrics, underscoring the need for cognitively grounded evaluation methods. Broadly, this work charts a path toward more reliable assessment frameworks for retrieval-augmented dialogue systems that better reflect the principles of natural human communication.
Hale, James; Dorley, Emmanuel J.; Kim, Peter H.; Gratch, Jonathan
IVAs to Study Gender Bias in Negotiation Journal Article
In: J Multimodal User Interfaces, vol. 20, no. 1, pp. 33–42, 2026, ISSN: 1783-7677, 1783-8738.
Abstract | Links | BibTeX | Tags: ARO, LLM, Social, Virtual Agents
@article{hale_ivas_2026,
title = {IVAs to Study Gender Bias in Negotiation},
author = {James Hale and Emmanuel J. Dorley and Peter H. Kim and Jonathan Gratch},
url = {https://link.springer.com/10.1007/s12193-026-00476-5},
doi = {10.1007/s12193-026-00476-5},
issn = {1783-7677, 1783-8738},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-18},
journal = {J Multimodal User Interfaces},
volume = {20},
number = {1},
pages = {33–42},
abstract = {Abstract
Virtual agents offer a powerful and underutilized methodology for studying social interaction, particularly how bias and stereotypes operate. Drawing on theoretical frameworks such as the Media Equation and supported by empirical studies, we demonstrate how virtual agents can serve as controlled proxies for human social actors to study social bias. Through two empirical studies in the negotiation domain, we show how gender cues can be studied in negotiation contexts. Study 1 examines the effects of gendered job descriptions on negotiation behavior — finding that gendered language and gender interact to affect participants’ minimally acceptable salary. Study 2 explores how agent embodiment and emotional tone influence user perceptions across gender lines — finding that agent embodiment and one’s gender interact to affect participants’ subjective feelings about the dispute. Together, these findings imply the need for greater use of virtual agents in research, as they help inform the development of more inclusive and equitable AI systems.},
keywords = {ARO, LLM, Social, Virtual Agents},
pubstate = {published},
tppubtype = {article}
}
Abstract
Virtual agents offer a powerful and underutilized methodology for studying social interaction, particularly how bias and stereotypes operate. Drawing on theoretical frameworks such as the Media Equation and supported by empirical studies, we demonstrate how virtual agents can serve as controlled proxies for human social actors to study social bias. Through two empirical studies in the negotiation domain, we show how gender cues can be studied in negotiation contexts. Study 1 examines the effects of gendered job descriptions on negotiation behavior — finding that gendered language and gender interact to affect participants’ minimally acceptable salary. Study 2 explores how agent embodiment and emotional tone influence user perceptions across gender lines — finding that agent embodiment and one’s gender interact to affect participants’ subjective feelings about the dispute. Together, these findings imply the need for greater use of virtual agents in research, as they help inform the development of more inclusive and equitable AI systems.
Virtual agents offer a powerful and underutilized methodology for studying social interaction, particularly how bias and stereotypes operate. Drawing on theoretical frameworks such as the Media Equation and supported by empirical studies, we demonstrate how virtual agents can serve as controlled proxies for human social actors to study social bias. Through two empirical studies in the negotiation domain, we show how gender cues can be studied in negotiation contexts. Study 1 examines the effects of gendered job descriptions on negotiation behavior — finding that gendered language and gender interact to affect participants’ minimally acceptable salary. Study 2 explores how agent embodiment and emotional tone influence user perceptions across gender lines — finding that agent embodiment and one’s gender interact to affect participants’ subjective feelings about the dispute. Together, these findings imply the need for greater use of virtual agents in research, as they help inform the development of more inclusive and equitable AI systems.
Hu, Yue; Wang, Daniel; Song, Xiyun; Lao, Zhiqiang; Zhou, Yubin; Chen, Meida; Feng, Andrew; Yu, Heather
Diffuse4D:Completing NeRF-Stereo Depth via Diffusion-Driven Restoration in Dynamic Scenes Proceedings Article
In: pp. 1003–1011, 2026.
@inproceedings{hu_diffuse4dcompleting_2026,
title = {Diffuse4D:Completing NeRF-Stereo Depth via Diffusion-Driven Restoration in Dynamic Scenes},
author = {Yue Hu and Daniel Wang and Xiyun Song and Zhiqiang Lao and Yubin Zhou and Meida Chen and Andrew Feng and Heather Yu},
year = {2026},
date = {2026-03-01},
pages = {1003–1011},
abstract = {4D reconstruction recovers time-varying 3D geometry, yet
accurate depth estimation in moving regions remains challenging. Most existing methods—from monocular predictors to 4D Gaussian Splatting—rely on single-view priors prone to scale-shift ambiguity and geometric distortion.
While stereo estimation offers metric accuracy through triangulation, NeRF-based stereo synthesis often introduces
ghosting and motion blur in dynamic areas, degrading performance. To address this, we propose a hybrid framework
combining NeRF-generated stereo depth with a confidence-
aware diffusion module to restore dynamic geometry. Our
approach eliminates affine uncertainty and monocular dependence, achieving temporally coherent sequences. Experimental results show our method reduces absolute relative error by over 50% and improves strict accuracy (δ <
1.05) by 10%, setting a new benchmark for high-fidelity 4D
reconstruction.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
4D reconstruction recovers time-varying 3D geometry, yet
accurate depth estimation in moving regions remains challenging. Most existing methods—from monocular predictors to 4D Gaussian Splatting—rely on single-view priors prone to scale-shift ambiguity and geometric distortion.
While stereo estimation offers metric accuracy through triangulation, NeRF-based stereo synthesis often introduces
ghosting and motion blur in dynamic areas, degrading performance. To address this, we propose a hybrid framework
combining NeRF-generated stereo depth with a confidence-
aware diffusion module to restore dynamic geometry. Our
approach eliminates affine uncertainty and monocular dependence, achieving temporally coherent sequences. Experimental results show our method reduces absolute relative error by over 50% and improves strict accuracy (δ <
1.05) by 10%, setting a new benchmark for high-fidelity 4D
reconstruction.
accurate depth estimation in moving regions remains challenging. Most existing methods—from monocular predictors to 4D Gaussian Splatting—rely on single-view priors prone to scale-shift ambiguity and geometric distortion.
While stereo estimation offers metric accuracy through triangulation, NeRF-based stereo synthesis often introduces
ghosting and motion blur in dynamic areas, degrading performance. To address this, we propose a hybrid framework
combining NeRF-generated stereo depth with a confidence-
aware diffusion module to restore dynamic geometry. Our
approach eliminates affine uncertainty and monocular dependence, achieving temporally coherent sequences. Experimental results show our method reduces absolute relative error by over 50% and improves strict accuracy (δ <
1.05) by 10%, setting a new benchmark for high-fidelity 4D
reconstruction.
O'Connell, Amy; Zhu, Xin; Kian, Mina; Dao, Warren; Gratch, Jonathan; Mataric, Maja; Culbertson, Heather
Exploring Remote Affective Communication Through a Haptic Wearable and Socially Assistive Robot Proceedings Article
In: Proceedings of the Twentieth International Conference on Tangible, Embedded, and Embodied Interaction, pp. 1–13, ACM, Chicago IL USA, 2026, ISBN: 979-8-4007-1868-7.
Links | BibTeX | Tags: Emotions, Social
@inproceedings{oconnell_exploring_2026,
title = {Exploring Remote Affective Communication Through a Haptic Wearable and Socially Assistive Robot},
author = {Amy O'Connell and Xin Zhu and Mina Kian and Warren Dao and Jonathan Gratch and Maja Mataric and Heather Culbertson},
url = {https://dl.acm.org/doi/10.1145/3731459.3773338},
doi = {10.1145/3731459.3773338},
isbn = {979-8-4007-1868-7},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-19},
booktitle = {Proceedings of the Twentieth International Conference on Tangible, Embedded, and Embodied Interaction},
pages = {1–13},
publisher = {ACM},
address = {Chicago IL USA},
keywords = {Emotions, Social},
pubstate = {published},
tppubtype = {inproceedings}
}
Tran, Minh; Siniukov, Maksim; Jin, Zhangyu; Soleymani, Mohammad
Discrete Facial Encoding: A Framework for the Data-driven Facial Display Discovery Proceedings Article
In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 2348–2358, 2026.
Abstract | BibTeX | Tags: ARO, DTIC?, Virtual Humans
@inproceedings{tran_discrete_2026,
title = {Discrete Facial Encoding: A Framework for the Data-driven Facial Display Discovery},
author = {Minh Tran and Maksim Siniukov and Zhangyu Jin and Mohammad Soleymani},
year = {2026},
date = {2026-03-01},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
pages = {2348–2358},
abstract = {Facial expression analysis is central to understanding human behavior, yet existing coding systems such as the Facial Action Coding System (FACS) are constrained by limited coverage and costly manual annotation. In this work,
we introduce Discrete Facial Encoding (DFE), an unsupervised, data-driven alternative of compact and interpretable dictionary of facial expressions from 3D mesh sequences learned through a Residual Vector Quantized Variational Autoencoder (RVQ-VAE). Our approach first extracts identityinvariant expression features from images using a 3D Morphable Model (3DMM), effectively disentangling factors such as head pose and facial geometry. We then encode these features using an RVQ-VAE, producing a sequence of discrete tokens from a shared codebook, where each token captures a specific, reusable facial deformation patter that contributes to the overall expression. Through extensive experiments, we demonstrate that Discrete Facial Encoding captures more precise facial behaviors than FACS
and other facial encoding alternatives. We evaluate the
utility of our representation across three high-level psychological tasks: stress detection, personality prediction, and depression detection. Using a simple Bag-of-Words model built on top of the learned tokens, our system consistently outperforms both FACS-based pipelines and strong image and video representation learning models such as Masked Autoencoders. Further analysis reveals that our representation covers a wider variety of facial displays, highlighting its potential as a scalable and effective alternative to FACS for psychological and affective computing applications. We released the source code and model weights at https://github.com/ihp-lab/vqface.},
keywords = {ARO, DTIC?, Virtual Humans},
pubstate = {published},
tppubtype = {inproceedings}
}
Facial expression analysis is central to understanding human behavior, yet existing coding systems such as the Facial Action Coding System (FACS) are constrained by limited coverage and costly manual annotation. In this work,
we introduce Discrete Facial Encoding (DFE), an unsupervised, data-driven alternative of compact and interpretable dictionary of facial expressions from 3D mesh sequences learned through a Residual Vector Quantized Variational Autoencoder (RVQ-VAE). Our approach first extracts identityinvariant expression features from images using a 3D Morphable Model (3DMM), effectively disentangling factors such as head pose and facial geometry. We then encode these features using an RVQ-VAE, producing a sequence of discrete tokens from a shared codebook, where each token captures a specific, reusable facial deformation patter that contributes to the overall expression. Through extensive experiments, we demonstrate that Discrete Facial Encoding captures more precise facial behaviors than FACS
and other facial encoding alternatives. We evaluate the
utility of our representation across three high-level psychological tasks: stress detection, personality prediction, and depression detection. Using a simple Bag-of-Words model built on top of the learned tokens, our system consistently outperforms both FACS-based pipelines and strong image and video representation learning models such as Masked Autoencoders. Further analysis reveals that our representation covers a wider variety of facial displays, highlighting its potential as a scalable and effective alternative to FACS for psychological and affective computing applications. We released the source code and model weights at https://github.com/ihp-lab/vqface.
we introduce Discrete Facial Encoding (DFE), an unsupervised, data-driven alternative of compact and interpretable dictionary of facial expressions from 3D mesh sequences learned through a Residual Vector Quantized Variational Autoencoder (RVQ-VAE). Our approach first extracts identityinvariant expression features from images using a 3D Morphable Model (3DMM), effectively disentangling factors such as head pose and facial geometry. We then encode these features using an RVQ-VAE, producing a sequence of discrete tokens from a shared codebook, where each token captures a specific, reusable facial deformation patter that contributes to the overall expression. Through extensive experiments, we demonstrate that Discrete Facial Encoding captures more precise facial behaviors than FACS
and other facial encoding alternatives. We evaluate the
utility of our representation across three high-level psychological tasks: stress detection, personality prediction, and depression detection. Using a simple Bag-of-Words model built on top of the learned tokens, our system consistently outperforms both FACS-based pipelines and strong image and video representation learning models such as Masked Autoencoders. Further analysis reveals that our representation covers a wider variety of facial displays, highlighting its potential as a scalable and effective alternative to FACS for psychological and affective computing applications. We released the source code and model weights at https://github.com/ihp-lab/vqface.
Xue, Jintang; Zhao, Ganning; Yao, Jie-En; Chen, Hong-En; Hu, Yue; Chen, Meida; You, Suya; Kuo, C. -C. Jay
Descrip3D: Enhancing Large Language Model-based 3D Scene Understanding with Object-Level Text Descriptions Proceedings Article
In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 1746–1756, 2026.
Abstract | BibTeX | Tags: LLM, Narrative
@inproceedings{xue_descrip3d_2026,
title = {Descrip3D: Enhancing Large Language Model-based 3D Scene Understanding with Object-Level Text Descriptions},
author = {Jintang Xue and Ganning Zhao and Jie-En Yao and Hong-En Chen and Yue Hu and Meida Chen and Suya You and C. -C. Jay Kuo},
year = {2026},
date = {2026-03-01},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
pages = {1746–1756},
abstract = {Understanding 3D scenes goes beyond simply recognizing objects; it requires reasoning about the spatial and semantic relationships between them. Current 3D scene-language models often struggle with this relational understanding, particularly when visual embeddings alone do not adequately convey the roles and interactions of objects. In this paper, we introduce Descrip3D, a novel and powerful framework that explicitly encodes the relationships between objects using natural language. Unlike previous methods that rely only on 2D and 3D embeddings, Descrip3D enhances each object with a textual description that captures both its intrinsic attributes and contextual relationships. These relational cues are incorporated into the model through a dual-level integration: embedding fusion and prompt-level injection. This allows for unified reasoning across various tasks such as grounding, captioning, and question answering, all without the need for task-specific heads or additional supervision. When evaluated on five benchmark datasets, including ScanRefer, Multi3DRefer, ScanQA, SQA3D, and Scan2Cap, Descrip3D consistently outperforms strong baseline models, demonstrating the effectiveness of language-guided relational representation for understanding complex indoor scenes.},
keywords = {LLM, Narrative},
pubstate = {published},
tppubtype = {inproceedings}
}
Understanding 3D scenes goes beyond simply recognizing objects; it requires reasoning about the spatial and semantic relationships between them. Current 3D scene-language models often struggle with this relational understanding, particularly when visual embeddings alone do not adequately convey the roles and interactions of objects. In this paper, we introduce Descrip3D, a novel and powerful framework that explicitly encodes the relationships between objects using natural language. Unlike previous methods that rely only on 2D and 3D embeddings, Descrip3D enhances each object with a textual description that captures both its intrinsic attributes and contextual relationships. These relational cues are incorporated into the model through a dual-level integration: embedding fusion and prompt-level injection. This allows for unified reasoning across various tasks such as grounding, captioning, and question answering, all without the need for task-specific heads or additional supervision. When evaluated on five benchmark datasets, including ScanRefer, Multi3DRefer, ScanQA, SQA3D, and Scan2Cap, Descrip3D consistently outperforms strong baseline models, demonstrating the effectiveness of language-guided relational representation for understanding complex indoor scenes.
Chaubey, Ashutosh; Guan, Xulang; Soleymani, Mohammad
Face-LLaVA: Facial Expression and Attribute Understanding through Instruction Tuning Proceedings Article
In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 2648–2660, 2026.
Abstract | BibTeX | Tags: ARO, DTIC, LLM, Natural Language, UARC
@inproceedings{chaubey_face-llava_2026,
title = {Face-LLaVA: Facial Expression and Attribute Understanding through Instruction Tuning},
author = {Ashutosh Chaubey and Xulang Guan and Mohammad Soleymani},
year = {2026},
date = {2026-03-01},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
pages = {2648–2660},
abstract = {The human face plays a central role in social communication, necessitating the use of performant computer vision tools for human-centered applications. We propose Face-LLaVA, a multimodal large language model for face-centered, in-context learning, including facial expression and attribute recognition. Additionally, Face-LLaVA is able to generate natural language descriptions that can be used for reasoning. Leveraging existing visual databases, we first developed FaceInstruct-1M, a face-centered database for instruction tuning MLLMs for face processing. We then developed a novel face-specific visual encoder powered by Face-Region Guided Cross-Attention that integrates face geometry with local visual features. We evaluated the proposed method across nine different datasets and five different face processing tasks, including facial expression recognition, action unit detection, facial attribute detection, age estimation and deepfake detection. Face-LLaVA achieves superior results compared to existing open-source MLLMs and competitive performance compared to commercial solutions. Our model output also receives a higher reasoning rating by GPT under a zero-shot setting across all the tasks. Both our dataset and model will be made public to support future advancements in social AI and foundational vision-language research.},
keywords = {ARO, DTIC, LLM, Natural Language, UARC},
pubstate = {published},
tppubtype = {inproceedings}
}
The human face plays a central role in social communication, necessitating the use of performant computer vision tools for human-centered applications. We propose Face-LLaVA, a multimodal large language model for face-centered, in-context learning, including facial expression and attribute recognition. Additionally, Face-LLaVA is able to generate natural language descriptions that can be used for reasoning. Leveraging existing visual databases, we first developed FaceInstruct-1M, a face-centered database for instruction tuning MLLMs for face processing. We then developed a novel face-specific visual encoder powered by Face-Region Guided Cross-Attention that integrates face geometry with local visual features. We evaluated the proposed method across nine different datasets and five different face processing tasks, including facial expression recognition, action unit detection, facial attribute detection, age estimation and deepfake detection. Face-LLaVA achieves superior results compared to existing open-source MLLMs and competitive performance compared to commercial solutions. Our model output also receives a higher reasoning rating by GPT under a zero-shot setting across all the tasks. Both our dataset and model will be made public to support future advancements in social AI and foundational vision-language research.
Jin, Zhangyu; Feng, Andrew; Chemburkar, Ankur; Melo, Celso M. De
PromptGAR: Flexible Promptive Group Activity Recognition Proceedings Article
In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pp. 4461–4471, 2026.
Abstract | Links | BibTeX | Tags: ARO, DTIC
@inproceedings{jin_promptgar_2026,
title = {PromptGAR: Flexible Promptive Group Activity Recognition},
author = {Zhangyu Jin and Andrew Feng and Ankur Chemburkar and Celso M. De Melo},
url = {https://openaccess.thecvf.com/content/WACV2026/papers/Jin_PromptGAR_Flexible_Promptive_Group_Activity_Recognition_WACV_2026_paper.pdf},
year = {2026},
date = {2026-03-01},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
pages = {4461–4471},
abstract = {We present PromptGAR, a novel framework for Group Activity Recognition (GAR) that offering both input flexibility and high recognition accuracy. The existing approaches suffer from limited real-world applicability due to their reliance on full prompt annotations, fixed number of frames and instances, and the lack of actor consistency. To bridge the gap, we proposed PromptGAR, which is the first GAR model to provide input flexibility across prompts, frames, and instances without the need for retraining. We leverage diverse visual prompts–like bounding boxes, skeletal keypoints, and instance identities–by unifying them as point prompts. A recognition decoder then cross-updates class and prompt tokens for enhanced performance. To ensure actor consistency for extended activity durations, we also introduce a relative instance attention mechanism that directly encodes instance identities. Comprehensive evaluations demonstrate that PromptGAR achieves competitive performances both on full prompts and partial prompt inputs, establishing its effectiveness on input flexibility and generalization ability for real-world applications.},
keywords = {ARO, DTIC},
pubstate = {published},
tppubtype = {inproceedings}
}
We present PromptGAR, a novel framework for Group Activity Recognition (GAR) that offering both input flexibility and high recognition accuracy. The existing approaches suffer from limited real-world applicability due to their reliance on full prompt annotations, fixed number of frames and instances, and the lack of actor consistency. To bridge the gap, we proposed PromptGAR, which is the first GAR model to provide input flexibility across prompts, frames, and instances without the need for retraining. We leverage diverse visual prompts–like bounding boxes, skeletal keypoints, and instance identities–by unifying them as point prompts. A recognition decoder then cross-updates class and prompt tokens for enhanced performance. To ensure actor consistency for extended activity durations, we also introduce a relative instance attention mechanism that directly encodes instance identities. Comprehensive evaluations demonstrate that PromptGAR achieves competitive performances both on full prompts and partial prompt inputs, establishing its effectiveness on input flexibility and generalization ability for real-world applications.
Liu, Ruying; Becerik-Gerber, Burcin; Lucas, Gale M.
In: International Journal of Human–Computer Interaction, pp. 1–20, 2026, ISSN: 1044-7318, 1532-7590.
Abstract | Links | BibTeX | Tags: ARO, DTIC?, VR
@article{liu_technology_2026,
title = {Technology Acceptance in High-Stakes Serious Games: Extending Technology Acceptance Model (TAM) with Affective and Experiential Predictors in Active Shooter Preparedness},
author = {Ruying Liu and Burcin Becerik-Gerber and Gale M. Lucas},
url = {https://www.tandfonline.com/doi/full/10.1080/10447318.2026.2635682},
doi = {10.1080/10447318.2026.2635682},
issn = {1044-7318, 1532-7590},
year = {2026},
date = {2026-03-01},
urldate = {2026-06-18},
journal = {International Journal of Human–Computer Interaction},
pages = {1–20},
abstract = {Active shooter incidents (ASIs) represent extreme threats to public safety. Virtual reality (VR) serious games offer promising training potential, but user acceptance in high-stakes and emotionally intense contexts is underexplored. This study extends the Technology Acceptance Model (TAM) by incorporating affective (perceived enjoyment) and experiential (self-reported engagement) factors. Using a desktop-based VR ASI preparedness simulation with 50 healthcare workers, we tested the extended TAM model. Core TAM relationships hold for ease of use and usefulness, but enjoyment and engagement play distinct roles. Enjoyment strongly predicted adoption, fully mediating the effect of ease of use and partially mediating usefulness. It was associated with high-arousal positive emotions(e.g., inspiration, pride, enthusiasm), reflecting eudaimonic states rather than hedonic fun. Engagement also significantly promoted adoption, though with weaker mediation effects. These findings extend TAM to high-stakes serious games and suggest that effective design should cultivate meaningful enjoyment and engagement without undermining scenario seriousness.},
keywords = {ARO, DTIC?, VR},
pubstate = {published},
tppubtype = {article}
}
Active shooter incidents (ASIs) represent extreme threats to public safety. Virtual reality (VR) serious games offer promising training potential, but user acceptance in high-stakes and emotionally intense contexts is underexplored. This study extends the Technology Acceptance Model (TAM) by incorporating affective (perceived enjoyment) and experiential (self-reported engagement) factors. Using a desktop-based VR ASI preparedness simulation with 50 healthcare workers, we tested the extended TAM model. Core TAM relationships hold for ease of use and usefulness, but enjoyment and engagement play distinct roles. Enjoyment strongly predicted adoption, fully mediating the effect of ease of use and partially mediating usefulness. It was associated with high-arousal positive emotions(e.g., inspiration, pride, enthusiasm), reflecting eudaimonic states rather than hedonic fun. Engagement also significantly promoted adoption, though with weaker mediation effects. These findings extend TAM to high-stakes serious games and suggest that effective design should cultivate meaningful enjoyment and engagement without undermining scenario seriousness.
Bodner, Ehud; Vol, Ilan; Bar-Natan, Dotan; Rizzo, Albert; Mikulincer, Mario; Maidenbaum, Shachar
Development and adaptation of virtual reality tools for PTSD and place-attachment treatment in Israel after the October 7th attacks Journal Article
In: Front. Virtual Real., vol. 7, pp. 1693453, 2026, ISSN: 2673-4192.
Abstract | Links | BibTeX | Tags: MedVR, VR
@article{bodner_development_2026,
title = {Development and adaptation of virtual reality tools for PTSD and place-attachment treatment in Israel after the October 7th attacks},
author = {Ehud Bodner and Ilan Vol and Dotan Bar-Natan and Albert Rizzo and Mario Mikulincer and Shachar Maidenbaum},
url = {https://www.frontiersin.org/articles/10.3389/frvir.2026.1693453/full},
doi = {10.3389/frvir.2026.1693453},
issn = {2673-4192},
year = {2026},
date = {2026-03-01},
urldate = {2026-05-21},
journal = {Front. Virtual Real.},
volume = {7},
pages = {1693453},
abstract = {The October 7 attacks caused significant psychological trauma among Israeli soldiers and civilians. Virtual reality (VR) has shown promise in PTSD treatment, particularly through BraveMind, a validated Prolonged Exposure (PE) VR system developed for U.S. veterans. This formative study examined how existing VR tools can be adapted to the Israeli context and extended beyond exposure to include place-attachment-based therapy for displaced civilians. Seven experienced clinicians participated in a focus-group evaluation of two systems: the original U.S.-based BraveMind and a newly developed place-attachment VR prototype (Re-PAVeR). Participants individually experienced brief VR exposures and completed questionnaires assessing presence, user experience, and perceived clinical utility, followed by a structured group discussion. Clinicians reported a strong sense of presence and positive attitudes toward VR-based interventions. They emphasized the importance of cultural, geographical, and operational adaptations to reflect Israeli combat and civilian trauma and highlighted the potential of attachment-based VR environments for addressing grief and loss among evacuees. Based on extensive prior validation of BraveMind and the present expert feedback, the findings support further development and contextual adaptation of VR-based interventions for PTSD treatment in Israel.},
keywords = {MedVR, VR},
pubstate = {published},
tppubtype = {article}
}
The October 7 attacks caused significant psychological trauma among Israeli soldiers and civilians. Virtual reality (VR) has shown promise in PTSD treatment, particularly through BraveMind, a validated Prolonged Exposure (PE) VR system developed for U.S. veterans. This formative study examined how existing VR tools can be adapted to the Israeli context and extended beyond exposure to include place-attachment-based therapy for displaced civilians. Seven experienced clinicians participated in a focus-group evaluation of two systems: the original U.S.-based BraveMind and a newly developed place-attachment VR prototype (Re-PAVeR). Participants individually experienced brief VR exposures and completed questionnaires assessing presence, user experience, and perceived clinical utility, followed by a structured group discussion. Clinicians reported a strong sense of presence and positive attitudes toward VR-based interventions. They emphasized the importance of cultural, geographical, and operational adaptations to reflect Israeli combat and civilian trauma and highlighted the potential of attachment-based VR environments for addressing grief and loss among evacuees. Based on extensive prior validation of BraveMind and the present expert feedback, the findings support further development and contextual adaptation of VR-based interventions for PTSD treatment in Israel.
Kwon, Deuksin; Shrestha, Kaleen; Han, Bin; Lin, Spencer; Hale, James; Gratch, Jonathan; Matarić, Maja; Lucas, Gale M.
Can LLMs Truly Embody Human Personality? Analyzing AI and Human Behavior Alignment in Dispute Resolution Miscellaneous
2026, (arXiv:2602.07414 [cs.AI]).
Abstract | Links | BibTeX | Tags: ARO, DTIC, LLM, Simulation, UARC
@misc{kwon_can_2026,
title = {Can LLMs Truly Embody Human Personality? Analyzing AI and Human Behavior Alignment in Dispute Resolution},
author = {Deuksin Kwon and Kaleen Shrestha and Bin Han and Spencer Lin and James Hale and Jonathan Gratch and Maja Matarić and Gale M. Lucas},
url = {http://arxiv.org/abs/2602.07414},
doi = {10.48550/arXiv.2602.07414},
year = {2026},
date = {2026-02-01},
urldate = {2026-06-18},
publisher = {arXiv},
abstract = {Large language models (LLMs) are increasingly used to simulate human behavior in social settings such as legal mediation, negotiation, and dispute resolution. However, it remains unclear whether these simulations reproduce the personality-behavior patterns observed in humans. Human personality, for instance, shapes how individuals navigate social interactions, including strategic choices and behaviors in emotionally charged interactions. This raises the question: Can LLMs, when prompted with personality traits, reproduce personality-driven differences in human conflict behavior? To explore this, we introduce an evaluation framework that enables direct comparison of human-human and LLM-LLM behaviors in dispute resolution dialogues with respect to Big Five Inventory (BFI) personality traits. This framework provides a set of interpretable metrics related to strategic behavior and conflict outcomes. We additionally contribute a novel dataset creation methodology for LLM dispute resolution dialogues with matched scenarios and personality traits with respect to human conversations. Finally, we demonstrate the use of our evaluation framework with three contemporary closed-source LLMs and show significant divergences in how personality manifests in conflict across different LLMs compared to human data, challenging the assumption that personality-prompted agents can serve as reliable behavioral proxies in socially impactful applications. Our work highlights the need for psychological grounding and validation in AI simulations before real-world use.},
note = {arXiv:2602.07414 [cs.AI]},
keywords = {ARO, DTIC, LLM, Simulation, UARC},
pubstate = {published},
tppubtype = {misc}
}
Large language models (LLMs) are increasingly used to simulate human behavior in social settings such as legal mediation, negotiation, and dispute resolution. However, it remains unclear whether these simulations reproduce the personality-behavior patterns observed in humans. Human personality, for instance, shapes how individuals navigate social interactions, including strategic choices and behaviors in emotionally charged interactions. This raises the question: Can LLMs, when prompted with personality traits, reproduce personality-driven differences in human conflict behavior? To explore this, we introduce an evaluation framework that enables direct comparison of human-human and LLM-LLM behaviors in dispute resolution dialogues with respect to Big Five Inventory (BFI) personality traits. This framework provides a set of interpretable metrics related to strategic behavior and conflict outcomes. We additionally contribute a novel dataset creation methodology for LLM dispute resolution dialogues with matched scenarios and personality traits with respect to human conversations. Finally, we demonstrate the use of our evaluation framework with three contemporary closed-source LLMs and show significant divergences in how personality manifests in conflict across different LLMs compared to human data, challenging the assumption that personality-prompted agents can serve as reliable behavioral proxies in socially impactful applications. Our work highlights the need for psychological grounding and validation in AI simulations before real-world use.
Chen, Pengcheng; Hu, Yue; Li, Wenhao; Gunderson, Nicole M.; Feng, Andrew; Sun, Zhenglong; Beerel, Peter; Seibel, Eric J.
NeVStereo: A NeRF-Driven NVS-Stereo Architecture for High-Fidelity 3D Tasks Miscellaneous
2026, (arXiv:2602.05423 [cs.CV]).
Abstract | Links | BibTeX | Tags: Geospatial, Graphics
@misc{chen_nevstereo_2026,
title = {NeVStereo: A NeRF-Driven NVS-Stereo Architecture for High-Fidelity 3D Tasks},
author = {Pengcheng Chen and Yue Hu and Wenhao Li and Nicole M. Gunderson and Andrew Feng and Zhenglong Sun and Peter Beerel and Eric J. Seibel},
url = {http://arxiv.org/abs/2602.05423},
doi = {10.48550/arXiv.2602.05423},
year = {2026},
date = {2026-02-01},
urldate = {2026-06-18},
publisher = {arXiv},
abstract = {In modern dense 3D reconstruction, feed-forward systems (e.g., VGGT, pi3) focus on end-to-end matching and geometry prediction but do not explicitly output the novel view synthesis (NVS). Neural rendering-based approaches offer high-fidelity NVS and detailed geometry from posed images, yet they typically assume fixed camera poses and can be sensitive to pose errors. As a result, it remains non-trivial to obtain a single framework that can offer accurate poses, reliable depth, high-quality rendering, and accurate 3D surfaces from casually captured views. We present NeVStereo, a NeRF-driven NVS-stereo architecture that aims to jointly deliver camera poses, multi-view depth, novel view synthesis, and surface reconstruction from multi-view RGB-only inputs. NeVStereo combines NeRF-based NVS for stereo-friendly renderings, confidence-guided multi-view depth estimation, NeRF-coupled bundle adjustment for pose refinement, and an iterative refinement stage that updates both depth and the radiance field to improve geometric consistency. This design mitigated the common NeRF-based issues such as surface stacking, artifacts, and pose-depth coupling. Across indoor, outdoor, tabletop, and aerial benchmarks, our experiments indicate that NeVStereo achieves consistently strong zero-shot performance, with up to 36% lower depth error, 10.4% improved pose accuracy, 4.5% higher NVS fidelity, and state-of-the-art mesh quality (F1 91.93%, Chamfer 4.35 mm) compared to existing prestigious methods.},
note = {arXiv:2602.05423 [cs.CV]},
keywords = {Geospatial, Graphics},
pubstate = {published},
tppubtype = {misc}
}
In modern dense 3D reconstruction, feed-forward systems (e.g., VGGT, pi3) focus on end-to-end matching and geometry prediction but do not explicitly output the novel view synthesis (NVS). Neural rendering-based approaches offer high-fidelity NVS and detailed geometry from posed images, yet they typically assume fixed camera poses and can be sensitive to pose errors. As a result, it remains non-trivial to obtain a single framework that can offer accurate poses, reliable depth, high-quality rendering, and accurate 3D surfaces from casually captured views. We present NeVStereo, a NeRF-driven NVS-stereo architecture that aims to jointly deliver camera poses, multi-view depth, novel view synthesis, and surface reconstruction from multi-view RGB-only inputs. NeVStereo combines NeRF-based NVS for stereo-friendly renderings, confidence-guided multi-view depth estimation, NeRF-coupled bundle adjustment for pose refinement, and an iterative refinement stage that updates both depth and the radiance field to improve geometric consistency. This design mitigated the common NeRF-based issues such as surface stacking, artifacts, and pose-depth coupling. Across indoor, outdoor, tabletop, and aerial benchmarks, our experiments indicate that NeVStereo achieves consistently strong zero-shot performance, with up to 36% lower depth error, 10.4% improved pose accuracy, 4.5% higher NVS fidelity, and state-of-the-art mesh quality (F1 91.93%, Chamfer 4.35 mm) compared to existing prestigious methods.
Han, Bin; Kwon, Deuksin; Gratch, Jonathan
Personality Expression Across Contexts: Linguistic and Behavioral Variation in LLM Agents Miscellaneous
2026, (arXiv:2602.01063 [cs.CL]).
Abstract | Links | BibTeX | Tags: LLM, Virtual Agents
@misc{han_personality_2026,
title = {Personality Expression Across Contexts: Linguistic and Behavioral Variation in LLM Agents},
author = {Bin Han and Deuksin Kwon and Jonathan Gratch},
url = {http://arxiv.org/abs/2602.01063},
doi = {10.48550/arXiv.2602.01063},
year = {2026},
date = {2026-02-01},
urldate = {2026-06-19},
publisher = {arXiv},
abstract = {Large Language Models (LLMs) can be conditioned with explicit personality prompts, yet their behavioral realization often varies depending on context. This study examines how identical personality prompts lead to distinct linguistic, behavioral, and emotional outcomes across four conversational settings: ice-breaking, negotiation, group decision, and empathy tasks. Results show that contextual cues systematically influence both personality expression and emotional tone, suggesting that the same traits are expressed differently depending on social and affective demands. This raises an important question for LLM-based dialogue agents: whether such variations reflect inconsistency or context-sensitive adaptation akin to human behavior. Viewed through the lens of Whole Trait Theory, these findings highlight that LLMs exhibit context-sensitive rather than fixed personality expression, adapting flexibly to social interaction goals and affective conditions.},
note = {arXiv:2602.01063 [cs.CL]},
keywords = {LLM, Virtual Agents},
pubstate = {published},
tppubtype = {misc}
}
Large Language Models (LLMs) can be conditioned with explicit personality prompts, yet their behavioral realization often varies depending on context. This study examines how identical personality prompts lead to distinct linguistic, behavioral, and emotional outcomes across four conversational settings: ice-breaking, negotiation, group decision, and empathy tasks. Results show that contextual cues systematically influence both personality expression and emotional tone, suggesting that the same traits are expressed differently depending on social and affective demands. This raises an important question for LLM-based dialogue agents: whether such variations reflect inconsistency or context-sensitive adaptation akin to human behavior. Viewed through the lens of Whole Trait Theory, these findings highlight that LLMs exhibit context-sensitive rather than fixed personality expression, adapting flexibly to social interaction goals and affective conditions.
Brizey, Jacqueline; Traum, David
Can code-switching improve the user experience with a dialogue system app for recording endangered languages? Proceedings Article
In: Proceedings of the 16th International Workshop on Spoken Dialogue Systems Technology, pp. 369–378, Association for Computational Linguistics, 2026.
Abstract | BibTeX | Tags: Dialogue
@inproceedings{brizey_can_2026,
title = {Can code-switching improve the user experience with a dialogue system app for recording endangered languages?},
author = {Jacqueline Brizey and David Traum},
year = {2026},
date = {2026-02-01},
booktitle = {Proceedings of the 16th International Workshop on Spoken Dialogue Systems Technology},
pages = {369–378},
publisher = {Association for Computational Linguistics},
abstract = {This paper investigates whether a multilingual
spoken dialogue system can be used to help
collect and preserve endangered language data.
In this work, we extend DAPEL (Dialogue
APp for Endangered Languages), which is designed to help preserve any language. Our focus, for testing purposes, is on the American Indigenous language Choctaw. The system uses
English as a common language, and we test
whether incorporating code-switching—the act
of alternating between languages—enhances
the user experience and/or increases the amount
of recorded language data. Our results indicate
that users have a positive response to interacting in both languages with the system, that the
system plays a meaningful role in language documentation, and, notably, that participants who
speak Choctaw as their first language are more
receptive to a code-switching system than to a
monolingual English-based system.},
keywords = {Dialogue},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper investigates whether a multilingual
spoken dialogue system can be used to help
collect and preserve endangered language data.
In this work, we extend DAPEL (Dialogue
APp for Endangered Languages), which is designed to help preserve any language. Our focus, for testing purposes, is on the American Indigenous language Choctaw. The system uses
English as a common language, and we test
whether incorporating code-switching—the act
of alternating between languages—enhances
the user experience and/or increases the amount
of recorded language data. Our results indicate
that users have a positive response to interacting in both languages with the system, that the
system plays a meaningful role in language documentation, and, notably, that participants who
speak Choctaw as their first language are more
receptive to a code-switching system than to a
monolingual English-based system.
spoken dialogue system can be used to help
collect and preserve endangered language data.
In this work, we extend DAPEL (Dialogue
APp for Endangered Languages), which is designed to help preserve any language. Our focus, for testing purposes, is on the American Indigenous language Choctaw. The system uses
English as a common language, and we test
whether incorporating code-switching—the act
of alternating between languages—enhances
the user experience and/or increases the amount
of recorded language data. Our results indicate
that users have a positive response to interacting in both languages with the system, that the
system plays a meaningful role in language documentation, and, notably, that participants who
speak Choctaw as their first language are more
receptive to a code-switching system than to a
monolingual English-based system.
Liu, Rong; Gao, Zhongpai; Planche, Benjamin; Chen, Meida; Nguyen, Van Nguyen; Zheng, Meng; Choudhuri, Anwesa; Chen, Terrence; Wang, Yue; Feng, Andrew; Wu, Ziyan
Universal Beta Splatting Miscellaneous
2026, (arXiv:2510.03312 [cs.GR]).
Abstract | Links | BibTeX | Tags: DTIC, Graphics, UARC
@misc{liu_universal_2026,
title = {Universal Beta Splatting},
author = {Rong Liu and Zhongpai Gao and Benjamin Planche and Meida Chen and Van Nguyen Nguyen and Meng Zheng and Anwesa Choudhuri and Terrence Chen and Yue Wang and Andrew Feng and Ziyan Wu},
url = {http://arxiv.org/abs/2510.03312},
doi = {10.48550/arXiv.2510.03312},
year = {2026},
date = {2026-02-01},
urldate = {2026-05-29},
publisher = {arXiv},
abstract = {We introduce Universal Beta Splatting (UBS), a unified framework that generalizes 3D Gaussian Splatting to N-dimensional anisotropic Beta kernels for explicit radiance field rendering. Unlike fixed Gaussian primitives, Beta kernels enable controllable dependency modeling across spatial, angular, and temporal dimensions within a single representation. Our unified approach captures complex light transport effects, handles anisotropic view-dependent appearance, and models scene dynamics without requiring auxiliary networks or specific color encodings. UBS maintains backward compatibility by approximating to Gaussian Splatting as a special case, guaranteeing plug-in usability and lower performance bounds. The learned Beta parameters naturally decompose scene properties into interpretable without explicit supervision: spatial (surface vs. texture), angular (diffuse vs. specular), and temporal (static vs. dynamic). Our CUDA-accelerated implementation achieves real-time rendering while consistently outperforming existing methods across static, view-dependent, and dynamic benchmarks, establishing Beta kernels as a scalable universal primitive for radiance field rendering. Our project website is available at https://rongliu-leo.github.io/universal-beta-splatting/.},
note = {arXiv:2510.03312 [cs.GR]},
keywords = {DTIC, Graphics, UARC},
pubstate = {published},
tppubtype = {misc}
}
We introduce Universal Beta Splatting (UBS), a unified framework that generalizes 3D Gaussian Splatting to N-dimensional anisotropic Beta kernels for explicit radiance field rendering. Unlike fixed Gaussian primitives, Beta kernels enable controllable dependency modeling across spatial, angular, and temporal dimensions within a single representation. Our unified approach captures complex light transport effects, handles anisotropic view-dependent appearance, and models scene dynamics without requiring auxiliary networks or specific color encodings. UBS maintains backward compatibility by approximating to Gaussian Splatting as a special case, guaranteeing plug-in usability and lower performance bounds. The learned Beta parameters naturally decompose scene properties into interpretable without explicit supervision: spatial (surface vs. texture), angular (diffuse vs. specular), and temporal (static vs. dynamic). Our CUDA-accelerated implementation achieves real-time rendering while consistently outperforming existing methods across static, view-dependent, and dynamic benchmarks, establishing Beta kernels as a scalable universal primitive for radiance field rendering. Our project website is available at https://rongliu-leo.github.io/universal-beta-splatting/.
Tak, Ala N.; Banayeeanzade, Amin; Bolourani, Anahita; Bahrani, Fatemeh; Chaubey, Ashutosh; Karimireddy, Sai Praneeth; Schwarz, Norbert; Gratch, Jonathan
Sparks of Rationality: Do Reasoning LLMs Align with Human Judgment and Choice? Miscellaneous
2026, (arXiv:2601.22329 [cs.AI]).
Abstract | Links | BibTeX | Tags: Emotions, LLM
@misc{tak_sparks_2026,
title = {Sparks of Rationality: Do Reasoning LLMs Align with Human Judgment and Choice?},
author = {Ala N. Tak and Amin Banayeeanzade and Anahita Bolourani and Fatemeh Bahrani and Ashutosh Chaubey and Sai Praneeth Karimireddy and Norbert Schwarz and Jonathan Gratch},
url = {http://arxiv.org/abs/2601.22329},
doi = {10.48550/arXiv.2601.22329},
year = {2026},
date = {2026-01-01},
urldate = {2026-06-19},
publisher = {arXiv},
abstract = {Large Language Models (LLMs) are increasingly positioned as decision engines for hiring, healthcare, and economic judgment, yet real-world human judgment reflects a balance between rational deliberation and emotion-driven bias. If LLMs are to participate in high-stakes decisions or serve as models of human behavior, it is critical to assess whether they exhibit analogous patterns of (ir)rationalities and biases. To this end, we evaluate multiple LLM families on (i) benchmarks testing core axioms of rational choice and (ii) classic decision domains from behavioral economics and social norms where emotions are known to shape judgment and choice. Across settings, we show that deliberate "thinking" reliably improves rationality and pushes models toward expected-value maximization. To probe human-like affective distortions and their interaction with reasoning, we use two emotion-steering methods: in-context priming (ICP) and representation-level steering (RLS). ICP induces strong directional shifts that are often extreme and difficult to calibrate, whereas RLS produces more psychologically plausible patterns but with lower reliability. Our results suggest that the same mechanisms that improve rationality also amplify sensitivity to affective interventions, and that different steering methods trade off controllability against human-aligned behavior. Overall, this points to a tension between reasoning and affective steering, with implications for both human simulation and the safe deployment of LLM-based decision systems.},
note = {arXiv:2601.22329 [cs.AI]},
keywords = {Emotions, LLM},
pubstate = {published},
tppubtype = {misc}
}
Large Language Models (LLMs) are increasingly positioned as decision engines for hiring, healthcare, and economic judgment, yet real-world human judgment reflects a balance between rational deliberation and emotion-driven bias. If LLMs are to participate in high-stakes decisions or serve as models of human behavior, it is critical to assess whether they exhibit analogous patterns of (ir)rationalities and biases. To this end, we evaluate multiple LLM families on (i) benchmarks testing core axioms of rational choice and (ii) classic decision domains from behavioral economics and social norms where emotions are known to shape judgment and choice. Across settings, we show that deliberate "thinking" reliably improves rationality and pushes models toward expected-value maximization. To probe human-like affective distortions and their interaction with reasoning, we use two emotion-steering methods: in-context priming (ICP) and representation-level steering (RLS). ICP induces strong directional shifts that are often extreme and difficult to calibrate, whereas RLS produces more psychologically plausible patterns but with lower reliability. Our results suggest that the same mechanisms that improve rationality also amplify sensitivity to affective interventions, and that different steering methods trade off controllability against human-aligned behavior. Overall, this points to a tension between reasoning and affective steering, with implications for both human simulation and the safe deployment of LLM-based decision systems.
Hale, James; Kim, HanMoe; Choi, Ahyoung; Kim, Peter H.; Gratch, Jonathan
Can LLMs Mediate Synchronous Dispute Dialogues Proceedings Article
In: pp. 117–131, CEUR Workshop Proceedings, 2026.
Abstract | Links | BibTeX | Tags: Dialogue, LLM
@inproceedings{hale_can_2026,
title = {Can LLMs Mediate Synchronous Dispute Dialogues},
author = {James Hale and HanMoe Kim and Ahyoung Choi and Peter H. Kim and Jonathan Gratch},
url = {https://ceur-ws.org/Vol-4178/paper8.pdf},
year = {2026},
date = {2026-01-01},
volume = {1},
pages = {117–131},
publisher = {CEUR Workshop Proceedings},
abstract = {Characterized by elevated hostility, disputes often result in disputants being unable to resolve their differences.
This emphasizes the role of a mediator; however, mediators are typically highly specialized, difficult to find,
and expensive. While alternatives exist, such as acquaintances or online moderators, these prove less effective.
This raises the question of whether AI can effectively facilitate contentious disputes. We assess the potential of
LLMs as mediators. First (Study 1), we focus on a large open-source corpus of customer service disputes that are
objectively classified in terms of two key reasons for mediation: (1) whether or not the dispute ended in success
or failure, and (2) to what extent the participants reported frustration with each other. We examine whether
LLMs could be prompted to predict when a mediator should intervene in advance, and show that the decision
to intervene is correctly sensitive to these two factors. Finally, we conducted a user study that compared AI to
human suggestions on when and how to intervene. We find that the LLMs were rated significantly better in
predicting when to intervene, rated as providing a better rationale for intervening, and rated as providing a more
effective mediation message to the disputants. Remarkably, observers preferred the AI 2-to-1 over the human
mediation decisions. Second (Study 2), we collect a small corpus of mediated dispute dialogues and analyze the
effectiveness of novice human mediators. Quantifying the similarity of novice mediators’ behaviors to those of
an LLM, we find the closer one mediates to an LLM, the better the outcome — both subjectively and objectively.
These results indicate LLMs not only outperform novice mediators but may serve to identify effective mediators.
While these results do not definitely demonstrate LLMs can mediate human disputes, we do show they can sense
escalation indicators and generate sensible messages — this work serves as a first-step toward AI-mediators in
human conflict.},
keywords = {Dialogue, LLM},
pubstate = {published},
tppubtype = {inproceedings}
}
Characterized by elevated hostility, disputes often result in disputants being unable to resolve their differences.
This emphasizes the role of a mediator; however, mediators are typically highly specialized, difficult to find,
and expensive. While alternatives exist, such as acquaintances or online moderators, these prove less effective.
This raises the question of whether AI can effectively facilitate contentious disputes. We assess the potential of
LLMs as mediators. First (Study 1), we focus on a large open-source corpus of customer service disputes that are
objectively classified in terms of two key reasons for mediation: (1) whether or not the dispute ended in success
or failure, and (2) to what extent the participants reported frustration with each other. We examine whether
LLMs could be prompted to predict when a mediator should intervene in advance, and show that the decision
to intervene is correctly sensitive to these two factors. Finally, we conducted a user study that compared AI to
human suggestions on when and how to intervene. We find that the LLMs were rated significantly better in
predicting when to intervene, rated as providing a better rationale for intervening, and rated as providing a more
effective mediation message to the disputants. Remarkably, observers preferred the AI 2-to-1 over the human
mediation decisions. Second (Study 2), we collect a small corpus of mediated dispute dialogues and analyze the
effectiveness of novice human mediators. Quantifying the similarity of novice mediators’ behaviors to those of
an LLM, we find the closer one mediates to an LLM, the better the outcome — both subjectively and objectively.
These results indicate LLMs not only outperform novice mediators but may serve to identify effective mediators.
While these results do not definitely demonstrate LLMs can mediate human disputes, we do show they can sense
escalation indicators and generate sensible messages — this work serves as a first-step toward AI-mediators in
human conflict.
This emphasizes the role of a mediator; however, mediators are typically highly specialized, difficult to find,
and expensive. While alternatives exist, such as acquaintances or online moderators, these prove less effective.
This raises the question of whether AI can effectively facilitate contentious disputes. We assess the potential of
LLMs as mediators. First (Study 1), we focus on a large open-source corpus of customer service disputes that are
objectively classified in terms of two key reasons for mediation: (1) whether or not the dispute ended in success
or failure, and (2) to what extent the participants reported frustration with each other. We examine whether
LLMs could be prompted to predict when a mediator should intervene in advance, and show that the decision
to intervene is correctly sensitive to these two factors. Finally, we conducted a user study that compared AI to
human suggestions on when and how to intervene. We find that the LLMs were rated significantly better in
predicting when to intervene, rated as providing a better rationale for intervening, and rated as providing a more
effective mediation message to the disputants. Remarkably, observers preferred the AI 2-to-1 over the human
mediation decisions. Second (Study 2), we collect a small corpus of mediated dispute dialogues and analyze the
effectiveness of novice human mediators. Quantifying the similarity of novice mediators’ behaviors to those of
an LLM, we find the closer one mediates to an LLM, the better the outcome — both subjectively and objectively.
These results indicate LLMs not only outperform novice mediators but may serve to identify effective mediators.
While these results do not definitely demonstrate LLMs can mediate human disputes, we do show they can sense
escalation indicators and generate sensible messages — this work serves as a first-step toward AI-mediators in
human conflict.
2025
Xiong, Haolin; Fu, Tianwen; Prasad, Pratusha Bhuvana; Cai, Yunxuan; Chen, Haiwei; Teng, Wenbin; Xiao, Hanyuan; Zhao, Yajie
Mind-to-Face: Neural-Driven Photorealistic Avatar Synthesis via EEG Decoding Miscellaneous
2025, (arXiv:2512.04313 [cs.CV]).
Abstract | Links | BibTeX | Tags: Emotions, VGL
@misc{xiong_mind–face_2025,
title = {Mind-to-Face: Neural-Driven Photorealistic Avatar Synthesis via EEG Decoding},
author = {Haolin Xiong and Tianwen Fu and Pratusha Bhuvana Prasad and Yunxuan Cai and Haiwei Chen and Wenbin Teng and Hanyuan Xiao and Yajie Zhao},
url = {http://arxiv.org/abs/2512.04313},
doi = {10.48550/arXiv.2512.04313},
year = {2025},
date = {2025-12-01},
urldate = {2026-06-22},
publisher = {arXiv},
abstract = {Current expressive avatar systems rely heavily on visual cues, failing when faces are occluded or when emotions remain internal. We present Mind-to-Face, the first framework that decodes non-invasive electroencephalogram (EEG) signals directly into high-fidelity facial expressions. We build a dual-modality recording setup to obtain synchronized EEG and multi-view facial video during emotion-eliciting stimuli, enabling precise supervision for neural-to-visual learning. Our model uses a CNN-Transformer encoder to map EEG signals into dense 3D position maps, capable of sampling over 65k vertices, capturing fine-scale geometry and subtle emotional dynamics, and renders them through a modified 3D Gaussian Splatting pipeline for photorealistic, view-consistent results. Through extensive evaluation, we show that EEG alone can reliably predict dynamic, subject-specific facial expressions, including subtle emotional responses, demonstrating that neural signals contain far richer affective and geometric information than previously assumed. Mind-to-Face establishes a new paradigm for neural-driven avatars, enabling personalized, emotion-aware telepresence and cognitive interaction in immersive environments.},
note = {arXiv:2512.04313 [cs.CV]},
keywords = {Emotions, VGL},
pubstate = {published},
tppubtype = {misc}
}
Current expressive avatar systems rely heavily on visual cues, failing when faces are occluded or when emotions remain internal. We present Mind-to-Face, the first framework that decodes non-invasive electroencephalogram (EEG) signals directly into high-fidelity facial expressions. We build a dual-modality recording setup to obtain synchronized EEG and multi-view facial video during emotion-eliciting stimuli, enabling precise supervision for neural-to-visual learning. Our model uses a CNN-Transformer encoder to map EEG signals into dense 3D position maps, capable of sampling over 65k vertices, capturing fine-scale geometry and subtle emotional dynamics, and renders them through a modified 3D Gaussian Splatting pipeline for photorealistic, view-consistent results. Through extensive evaluation, we show that EEG alone can reliably predict dynamic, subject-specific facial expressions, including subtle emotional responses, demonstrating that neural signals contain far richer affective and geometric information than previously assumed. Mind-to-Face establishes a new paradigm for neural-driven avatars, enabling personalized, emotion-aware telepresence and cognitive interaction in immersive environments.
Rodrigues, Patrick Borges; Lucas, Gale M.; Fang, Yiyang; Wang, Zihao; Roll, Shawn C.; Becerik-Gerber, Burcin; Soibelman, Lucio
Understanding Operators’ Sensory Needs for Human–Robot Interaction in Teleoperated Demolition Journal Article
In: J. Constr. Eng. Manage., vol. 151, no. 12, pp. 04025207, 2025, ISSN: 0733-9364, 1943-7862.
Abstract | Links | BibTeX | Tags: ARO
@article{rodrigues_understanding_2025-1,
title = {Understanding Operators’ Sensory Needs for Human–Robot Interaction in Teleoperated Demolition},
author = {Patrick Borges Rodrigues and Gale M. Lucas and Yiyang Fang and Zihao Wang and Shawn C. Roll and Burcin Becerik-Gerber and Lucio Soibelman},
url = {https://ascelibrary.org/doi/10.1061/JCEMD4.COENG-16911},
doi = {10.1061/JCEMD4.COENG-16911},
issn = {0733-9364, 1943-7862},
year = {2025},
date = {2025-12-01},
urldate = {2026-06-22},
journal = {J. Constr. Eng. Manage.},
volume = {151},
number = {12},
pages = {04025207},
abstract = {Despite the benefits of remotely operated compact demolition machines in constrained and hazardous environments, many safety and occupational risks remain. Operators often stay close to the machines, exposing themselves to the risks of structural collapses, falling debris, dust, and noise. In this study, we employed a generative user experience (UX) framework with six demolition industry professionals to inform the design considerations for a sensory-enhanced teleoperation interface for demolition robots. Through narrative interviews and focus groups, we identified demolition operators’ sensory needs during traditional demolition tasks involving manned and remotely operated machines and explored potential sensory features for a teleoperation interface for demolition robots. To increase the trustworthiness of the findings, we subsequently conducted a verification strategy where over 60 demolition stakeholders tested and provided feedback on the proposed teleoperation interface features within a virtual environment. Findings indicate that demolition operators use multiple sensory channels to perceive and interact with the environment during traditional demolition tasks. However, certain site conditions (e.g., noise, constrained spaces, clutter) still compromise their performance, safety, and well-being. Multimodal, simple, safe, and personalizable teleoperation interfaces that provide appropriate levels of control over the robots are preferred. Key suggestions included improvements in the robot’s controllers and the inclusion of mechanisms to facilitate communications with on-site workers and the perception of hazards. The findings inform the design considerations for a sensory-enhanced teleoperation interface for compact demolition robots. Additionally, the proposed generative UX framework can be applied to other construction applications to facilitate identifying operators’ needs and designing teleoperation interfaces.},
keywords = {ARO},
pubstate = {published},
tppubtype = {article}
}
Despite the benefits of remotely operated compact demolition machines in constrained and hazardous environments, many safety and occupational risks remain. Operators often stay close to the machines, exposing themselves to the risks of structural collapses, falling debris, dust, and noise. In this study, we employed a generative user experience (UX) framework with six demolition industry professionals to inform the design considerations for a sensory-enhanced teleoperation interface for demolition robots. Through narrative interviews and focus groups, we identified demolition operators’ sensory needs during traditional demolition tasks involving manned and remotely operated machines and explored potential sensory features for a teleoperation interface for demolition robots. To increase the trustworthiness of the findings, we subsequently conducted a verification strategy where over 60 demolition stakeholders tested and provided feedback on the proposed teleoperation interface features within a virtual environment. Findings indicate that demolition operators use multiple sensory channels to perceive and interact with the environment during traditional demolition tasks. However, certain site conditions (e.g., noise, constrained spaces, clutter) still compromise their performance, safety, and well-being. Multimodal, simple, safe, and personalizable teleoperation interfaces that provide appropriate levels of control over the robots are preferred. Key suggestions included improvements in the robot’s controllers and the inclusion of mechanisms to facilitate communications with on-site workers and the perception of hazards. The findings inform the design considerations for a sensory-enhanced teleoperation interface for compact demolition robots. Additionally, the proposed generative UX framework can be applied to other construction applications to facilitate identifying operators’ needs and designing teleoperation interfaces.
Chaudhri, Vinay K; Baru, Chaitan; Bennett, Brandon; Bhatt, Mehul; Cassel, Darion; Cohn, Anthony G; Dechter, Rina; Erdem, Esra; Ferrucci, Dave; Forbus, Ken; Gelfond, Gregory; Genesereth, Michael; Gordon, Andrew S.; Grosof, Benjamin; Gupta, Gopal; Hendler, Jim; Israni, Sharat; Josephson, Tyler R.; Kyllonen, Patrick; Lierler, Yuliya; Lifschitz, Vladimir; McFate, Clifton; McGinty, Hande Küçük; Morgenstern, Leora; Oltramari, Alessandro; Paritosh, Praveen; Roth, Dan; Shepard, Blake; Shimizu, Cogan; Vrandečić, Denny; Whiting, Mark; Witbrock, Michael
A community‐driven vision for a new knowledge resource for AI Journal Article
In: AI Magazine, vol. 46, no. 4, pp. e70035, 2025, ISSN: 0738-4602, 2371-9621.
Abstract | Links | BibTeX | Tags: AI, LLM
@article{chaudhri_communitydriven_2025,
title = {A community‐driven vision for a new knowledge resource for AI},
author = {Vinay K Chaudhri and Chaitan Baru and Brandon Bennett and Mehul Bhatt and Darion Cassel and Anthony G Cohn and Rina Dechter and Esra Erdem and Dave Ferrucci and Ken Forbus and Gregory Gelfond and Michael Genesereth and Andrew S. Gordon and Benjamin Grosof and Gopal Gupta and Jim Hendler and Sharat Israni and Tyler R. Josephson and Patrick Kyllonen and Yuliya Lierler and Vladimir Lifschitz and Clifton McFate and Hande Küçük McGinty and Leora Morgenstern and Alessandro Oltramari and Praveen Paritosh and Dan Roth and Blake Shepard and Cogan Shimizu and Denny Vrandečić and Mark Whiting and Michael Witbrock},
url = {https://onlinelibrary.wiley.com/doi/10.1002/aaai.70035},
doi = {10.1002/aaai.70035},
issn = {0738-4602, 2371-9621},
year = {2025},
date = {2025-12-01},
urldate = {2026-06-23},
journal = {AI Magazine},
volume = {46},
number = {4},
pages = {e70035},
abstract = {Abstract
The long‐standing goal of creating a comprehensive, multi‐purpose knowledge resource, reminiscent of the 1984 Cyc project, still persists in AI. Despite the success of knowledge resources like WordNet, ConceptNet, WolframtextbarAlpha and other commercial knowledge graphs, verifiable, general‐purpose, widely available sources of knowledge remain a critical deficiency in AI infrastructure. Large language models struggle due to knowledge gaps; robotic planning lacks necessary world knowledge; and the detection of factually false information relies heavily on human expertise. What kind of knowledge resource is most needed in AI today? How can modern technology shape its development and evaluation? A recent AAAI workshop gathered over 50 researchers to explore these questions. This paper synthesizes our findings and outlines a community‐driven vision for a new knowledge infrastructure. In addition to leveraging contemporary advances in knowledge representation and reasoning, one promising idea is to build an open engineering framework to exploit knowledge modules effectively within the context of practical applications. Such a framework should include sets of conventions and social structures that are adopted by contributors.},
keywords = {AI, LLM},
pubstate = {published},
tppubtype = {article}
}
Abstract
The long‐standing goal of creating a comprehensive, multi‐purpose knowledge resource, reminiscent of the 1984 Cyc project, still persists in AI. Despite the success of knowledge resources like WordNet, ConceptNet, WolframtextbarAlpha and other commercial knowledge graphs, verifiable, general‐purpose, widely available sources of knowledge remain a critical deficiency in AI infrastructure. Large language models struggle due to knowledge gaps; robotic planning lacks necessary world knowledge; and the detection of factually false information relies heavily on human expertise. What kind of knowledge resource is most needed in AI today? How can modern technology shape its development and evaluation? A recent AAAI workshop gathered over 50 researchers to explore these questions. This paper synthesizes our findings and outlines a community‐driven vision for a new knowledge infrastructure. In addition to leveraging contemporary advances in knowledge representation and reasoning, one promising idea is to build an open engineering framework to exploit knowledge modules effectively within the context of practical applications. Such a framework should include sets of conventions and social structures that are adopted by contributors.
The long‐standing goal of creating a comprehensive, multi‐purpose knowledge resource, reminiscent of the 1984 Cyc project, still persists in AI. Despite the success of knowledge resources like WordNet, ConceptNet, WolframtextbarAlpha and other commercial knowledge graphs, verifiable, general‐purpose, widely available sources of knowledge remain a critical deficiency in AI infrastructure. Large language models struggle due to knowledge gaps; robotic planning lacks necessary world knowledge; and the detection of factually false information relies heavily on human expertise. What kind of knowledge resource is most needed in AI today? How can modern technology shape its development and evaluation? A recent AAAI workshop gathered over 50 researchers to explore these questions. This paper synthesizes our findings and outlines a community‐driven vision for a new knowledge infrastructure. In addition to leveraging contemporary advances in knowledge representation and reasoning, one promising idea is to build an open engineering framework to exploit knowledge modules effectively within the context of practical applications. Such a framework should include sets of conventions and social structures that are adopted by contributors.
Klapow, Maxwell; Seeff, Misha; Adam, Maya; Greuel, Merlin; Hoffman, Daniel; Rogge, Jessica R.; Gordon, Andrew; Bärnighausen, Till; Amsalem, Doron
A short, animated storytelling video to reduce addiction stigma: A pilot randomized controlled trial Journal Article
In: Addictive Behaviors Reports, vol. 22, pp. 100622, 2025, ISSN: 23528532.
Abstract | Links | BibTeX | Tags: The Narrative Group
@article{klapow_short_2025,
title = {A short, animated storytelling video to reduce addiction stigma: A pilot randomized controlled trial},
author = {Maxwell Klapow and Misha Seeff and Maya Adam and Merlin Greuel and Daniel Hoffman and Jessica R. Rogge and Andrew Gordon and Till Bärnighausen and Doron Amsalem},
url = {https://linkinghub.elsevier.com/retrieve/pii/S2352853225000409},
doi = {10.1016/j.abrep.2025.100622},
issn = {23528532},
year = {2025},
date = {2025-12-01},
urldate = {2026-06-25},
journal = {Addictive Behaviors Reports},
volume = {22},
pages = {100622},
abstract = {Background and aims: Public stigma towards people with addiction negatively impacts help-seeking, treatment
and recovery. This pilot study tested the feasibility of conducting a large-scale, online trial to measure the effect
of a short, animated storytelling (SAS) stigma reduction video, with and without soundtrack, on addiction
stigma, optimism, warmth towards people with addiction, and hopefulness at two timepoints (immediately postexposure and 14 days later).
Design: We used a parallel group, three-arm randomized controlled trial (RCT).
Setting: We conducted this fully online study on the Prolific Academic research platform (participant recruitment)
and the Qualtrics survey platform (data collection).
Participants: We recruited 631 English-speaking adult participants, aged 18–49, residing in the United States.
Interventions: Intervention group A received the SAS video intervention. Intervention group B group received the
SAS video intervention without sound. The control group received written information about global addiction
prevalence, estimated to be time-equivalent with the video interventions.
Measurements: We measured participant retention rate at the two-week follow-up to determine the feasibility of
conducting the definitive trial. Our co-primary outcomes were addiction stigma, optimism, warmth towards
people with addiction and hopefulness, measured using an abbreviated 18-item version of the Attribution
Questionnaire (AQ-18), the Brief García’s Interactive Optimism Scale (BIOS-G), a stigma thermometer and a
visual analogue scale (VAS). We used repeated-measures ANOVA to assess group-by-time interactions and
compared changes from baseline to post-intervention. Participants completed follow-up surveys 14 days postintervention.
Findings: The retention rate from baseline to follow-up was 88.0 %. Exposure to both the video with sound and without sound resulted in significant positive changes compared to the control group, for pity [F (4,1046) = 3.26, η2 = 0.012},
keywords = {The Narrative Group},
pubstate = {published},
tppubtype = {article}
}
Background and aims: Public stigma towards people with addiction negatively impacts help-seeking, treatment
and recovery. This pilot study tested the feasibility of conducting a large-scale, online trial to measure the effect
of a short, animated storytelling (SAS) stigma reduction video, with and without soundtrack, on addiction
stigma, optimism, warmth towards people with addiction, and hopefulness at two timepoints (immediately postexposure and 14 days later).
Design: We used a parallel group, three-arm randomized controlled trial (RCT).
Setting: We conducted this fully online study on the Prolific Academic research platform (participant recruitment)
and the Qualtrics survey platform (data collection).
Participants: We recruited 631 English-speaking adult participants, aged 18–49, residing in the United States.
Interventions: Intervention group A received the SAS video intervention. Intervention group B group received the
SAS video intervention without sound. The control group received written information about global addiction
prevalence, estimated to be time-equivalent with the video interventions.
Measurements: We measured participant retention rate at the two-week follow-up to determine the feasibility of
conducting the definitive trial. Our co-primary outcomes were addiction stigma, optimism, warmth towards
people with addiction and hopefulness, measured using an abbreviated 18-item version of the Attribution
Questionnaire (AQ-18), the Brief García’s Interactive Optimism Scale (BIOS-G), a stigma thermometer and a
visual analogue scale (VAS). We used repeated-measures ANOVA to assess group-by-time interactions and
compared changes from baseline to post-intervention. Participants completed follow-up surveys 14 days postintervention.
Findings: The retention rate from baseline to follow-up was 88.0 %. Exposure to both the video with sound and without sound resulted in significant positive changes compared to the control group, for pity [F (4,1046) = 3.26, η2 = 0.012
and recovery. This pilot study tested the feasibility of conducting a large-scale, online trial to measure the effect
of a short, animated storytelling (SAS) stigma reduction video, with and without soundtrack, on addiction
stigma, optimism, warmth towards people with addiction, and hopefulness at two timepoints (immediately postexposure and 14 days later).
Design: We used a parallel group, three-arm randomized controlled trial (RCT).
Setting: We conducted this fully online study on the Prolific Academic research platform (participant recruitment)
and the Qualtrics survey platform (data collection).
Participants: We recruited 631 English-speaking adult participants, aged 18–49, residing in the United States.
Interventions: Intervention group A received the SAS video intervention. Intervention group B group received the
SAS video intervention without sound. The control group received written information about global addiction
prevalence, estimated to be time-equivalent with the video interventions.
Measurements: We measured participant retention rate at the two-week follow-up to determine the feasibility of
conducting the definitive trial. Our co-primary outcomes were addiction stigma, optimism, warmth towards
people with addiction and hopefulness, measured using an abbreviated 18-item version of the Attribution
Questionnaire (AQ-18), the Brief García’s Interactive Optimism Scale (BIOS-G), a stigma thermometer and a
visual analogue scale (VAS). We used repeated-measures ANOVA to assess group-by-time interactions and
compared changes from baseline to post-intervention. Participants completed follow-up surveys 14 days postintervention.
Findings: The retention rate from baseline to follow-up was 88.0 %. Exposure to both the video with sound and without sound resulted in significant positive changes compared to the control group, for pity [F (4,1046) = 3.26, η2 = 0.012
Okado, Yuko; Nye, Benjamin D.; Aguirre, Angelica; Swartout, William
In: International Journal of Artificial Intelligence in Education, vol. 35, no. 4, pp. 2596–2615, 2025, ISSN: 15604292.
Abstract | Links | BibTeX | Tags: Dialogue, Virtual Agents
@article{okado_how_2025-1,
title = {How Can Virtual Agents Scale Up Mentoring?: Insights from College Students’ Experiences Using the CareerFair.ai Platform at an American Hispanic-Serving Institution},
author = {Yuko Okado and Benjamin D. Nye and Angelica Aguirre and William Swartout},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1560429226000491},
doi = {10.1007/s40593-025-00482-w},
issn = {15604292},
year = {2025},
date = {2025-12-01},
urldate = {2026-06-25},
journal = {International Journal of Artificial Intelligence in Education},
volume = {35},
number = {4},
pages = {2596–2615},
abstract = {Mentoring promotes underserved students’ persistence in STEM but is difficult to scale up. Conversational virtual agents can help address this problem by conveying a mentor’s experiences to larger audiences. The present study examined college students’ (N = 138) utilization of CareerFair.ai, an online platform featuring virtual agent-mentors that were self-recorded by sixteen real-life mentors and built using principles from the earlier MentorPal framework. Participants completed a single-session study which included 30 minutes of active interaction with CareerFair.ai, sandwiched between pre-test and post-test surveys. Students’ user experience and learning gains were examined, both for the overall sample and with a lens of diversity and equity across different, potentially underserved demographic groups. Findings included positive pre/post changes in intent to pursue STEM coursework and high user acceptance ratings (e.g., expected benefit, ease of use), with under-represented minority (URM) students giving significantly higher ratings on average than non-URM students. Self-reported learning gains of interest, actual content viewed on the CareerFair.ai platform, and actual learning gains were associated with one another, suggesting that the platform may be a useful resource in meeting a wide range of career exploration needs. Overall, the CareerFair.ai platform shows promise in scaling up aspects of mentoring to serve the needs of diverse groups of college students.},
keywords = {Dialogue, Virtual Agents},
pubstate = {published},
tppubtype = {article}
}
Mentoring promotes underserved students’ persistence in STEM but is difficult to scale up. Conversational virtual agents can help address this problem by conveying a mentor’s experiences to larger audiences. The present study examined college students’ (N = 138) utilization of CareerFair.ai, an online platform featuring virtual agent-mentors that were self-recorded by sixteen real-life mentors and built using principles from the earlier MentorPal framework. Participants completed a single-session study which included 30 minutes of active interaction with CareerFair.ai, sandwiched between pre-test and post-test surveys. Students’ user experience and learning gains were examined, both for the overall sample and with a lens of diversity and equity across different, potentially underserved demographic groups. Findings included positive pre/post changes in intent to pursue STEM coursework and high user acceptance ratings (e.g., expected benefit, ease of use), with under-represented minority (URM) students giving significantly higher ratings on average than non-URM students. Self-reported learning gains of interest, actual content viewed on the CareerFair.ai platform, and actual learning gains were associated with one another, suggesting that the platform may be a useful resource in meeting a wide range of career exploration needs. Overall, the CareerFair.ai platform shows promise in scaling up aspects of mentoring to serve the needs of diverse groups of college students.
Xu, Ziyue; Zhang, Zhihong; Roth, Holger R.; Chen, Chester; Cheng, Yan; Feng, Andrew
Optimizing Federated Learning in the Era of LLMs: Message Quantization and Streaming Miscellaneous
2025, (arXiv:2511.16450 [cs.DC]).
Abstract | Links | BibTeX | Tags: LLM, Machine Learning
@misc{xu_optimizing_2025,
title = {Optimizing Federated Learning in the Era of LLMs: Message Quantization and Streaming},
author = {Ziyue Xu and Zhihong Zhang and Holger R. Roth and Chester Chen and Yan Cheng and Andrew Feng},
url = {http://arxiv.org/abs/2511.16450},
doi = {10.48550/arXiv.2511.16450},
year = {2025},
date = {2025-11-01},
urldate = {2026-06-25},
publisher = {arXiv},
abstract = {Federated Learning (FL) offers a promising solution for training machine learning models across distributed data sources while preserving data privacy. However, FL faces critical challenges related to communication overhead and local resource constraints, especially in the era of Large Language Models (LLMs) with billions of parameters. The sheer size of these models exacerbates both memory and communication constraints, making efficient transmission and processing essential for practical deployment. NVIDIA FLARE, an open-source SDK for federated learning, addresses these challenges by introducing advanced communication capabilities. Building upon existing solutions for large object streaming, we enhance FL workflows for LLMs through two key techniques: message quantization and container/file streaming. Quantization reduces message size, while streaming enables efficient memory management, improving scalability and integration with existing workflows. These advancements significantly enhance the robustness and efficiency of FL with LLMs, ensuring better performance in real-world federated learning scenarios.},
note = {arXiv:2511.16450 [cs.DC]},
keywords = {LLM, Machine Learning},
pubstate = {published},
tppubtype = {misc}
}
Federated Learning (FL) offers a promising solution for training machine learning models across distributed data sources while preserving data privacy. However, FL faces critical challenges related to communication overhead and local resource constraints, especially in the era of Large Language Models (LLMs) with billions of parameters. The sheer size of these models exacerbates both memory and communication constraints, making efficient transmission and processing essential for practical deployment. NVIDIA FLARE, an open-source SDK for federated learning, addresses these challenges by introducing advanced communication capabilities. Building upon existing solutions for large object streaming, we enhance FL workflows for LLMs through two key techniques: message quantization and container/file streaming. Quantization reduces message size, while streaming enables efficient memory management, improving scalability and integration with existing workflows. These advancements significantly enhance the robustness and efficiency of FL with LLMs, ensuring better performance in real-world federated learning scenarios.
Kwon, Deuksin; Hae, Jiwon; Clift, Emma; Shamsoddini, Daniel; Gratch, Jonathan; Lucas, Gale
ASTRA: A Negotiation Agent with Adaptive and Strategic Reasoning through Action in Dynamic Offer Optimization Proceedings Article
In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pp. 16217–16238, Association for Computational Linguistics, 2025.
Abstract | Links | BibTeX | Tags:
@inproceedings{kwon_astra_2025-1,
title = {ASTRA: A Negotiation Agent with Adaptive and Strategic Reasoning through Action in Dynamic Offer Optimization},
author = {Deuksin Kwon and Jiwon Hae and Emma Clift and Daniel Shamsoddini and Jonathan Gratch and Gale Lucas},
url = {aclanthology.org/2025.emnlp-main.821.pdf},
year = {2025},
date = {2025-11-01},
booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
pages = {16217–16238},
publisher = {Association for Computational Linguistics},
abstract = {Negotiation requires dynamically balancing
self-interest and cooperation within the flow
of conversation to maximize one’s own utility.
Yet, existing agents struggle due to bounded
rationality in human data, low adaptability
to counterpart behavior, and limited strategic reasoning. To address this, we introduce
principle-driven negotiation agents, powered
by ASTRA, a novel framework for turn-level
offer optimization grounded in two core principles: opponent modeling and Tit-for-Tat reciprocity. ASTRA operates in three stages: (1)
interpreting counterpart behavior, (2) optimizing counteroffers via a tool-integrated action
with a linear programming (LP) solver, and
(3) selecting offers based on strategy assessment and the partner’s acceptance probability. Through simulations and human evaluations, our agent effectively adapts to an opponent’s shifting stance and achieves favorable
outcomes through enhanced adaptability and
strategic reasoning. Beyond enhancing negotiation performance, it also serves as a powerful
coaching tool, offering interpretable strategic
feedback and optimal offer recommendations
beyond human bounded rationality, with its potential further validated through human evaluation.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Negotiation requires dynamically balancing
self-interest and cooperation within the flow
of conversation to maximize one’s own utility.
Yet, existing agents struggle due to bounded
rationality in human data, low adaptability
to counterpart behavior, and limited strategic reasoning. To address this, we introduce
principle-driven negotiation agents, powered
by ASTRA, a novel framework for turn-level
offer optimization grounded in two core principles: opponent modeling and Tit-for-Tat reciprocity. ASTRA operates in three stages: (1)
interpreting counterpart behavior, (2) optimizing counteroffers via a tool-integrated action
with a linear programming (LP) solver, and
(3) selecting offers based on strategy assessment and the partner’s acceptance probability. Through simulations and human evaluations, our agent effectively adapts to an opponent’s shifting stance and achieves favorable
outcomes through enhanced adaptability and
strategic reasoning. Beyond enhancing negotiation performance, it also serves as a powerful
coaching tool, offering interpretable strategic
feedback and optimal offer recommendations
beyond human bounded rationality, with its potential further validated through human evaluation.
self-interest and cooperation within the flow
of conversation to maximize one’s own utility.
Yet, existing agents struggle due to bounded
rationality in human data, low adaptability
to counterpart behavior, and limited strategic reasoning. To address this, we introduce
principle-driven negotiation agents, powered
by ASTRA, a novel framework for turn-level
offer optimization grounded in two core principles: opponent modeling and Tit-for-Tat reciprocity. ASTRA operates in three stages: (1)
interpreting counterpart behavior, (2) optimizing counteroffers via a tool-integrated action
with a linear programming (LP) solver, and
(3) selecting offers based on strategy assessment and the partner’s acceptance probability. Through simulations and human evaluations, our agent effectively adapts to an opponent’s shifting stance and achieves favorable
outcomes through enhanced adaptability and
strategic reasoning. Beyond enhancing negotiation performance, it also serves as a powerful
coaching tool, offering interpretable strategic
feedback and optimal offer recommendations
beyond human bounded rationality, with its potential further validated through human evaluation.
Hans, Soham; Ustun, Volkan; Nye, Benjamin; Sterrett, James; Green, Matthew
Towards AI-Assisted Generation of Military Training Scenarios Miscellaneous
2025, (arXiv:2511.07690 [cs.AI]).
Abstract | Links | BibTeX | Tags: AI, DTIC, Learning Sciences, LLM, UARC
@misc{hans_towards_2025,
title = {Towards AI-Assisted Generation of Military Training Scenarios},
author = {Soham Hans and Volkan Ustun and Benjamin Nye and James Sterrett and Matthew Green},
url = {http://arxiv.org/abs/2511.07690},
doi = {10.48550/arXiv.2511.07690},
year = {2025},
date = {2025-11-01},
urldate = {2026-06-17},
publisher = {arXiv},
abstract = {Achieving expert-level performance in simulation-based training relies on the creation of complex, adaptable scenarios, a traditionally laborious and resource intensive process. Although prior research explored scenario generation for military training, pre-LLM AI tools struggled to generate sufficiently complex or adaptable scenarios. This paper introduces a multi-agent, multi-modal reasoning framework that leverages Large Language Models (LLMs) to generate critical training artifacts, such as Operations Orders (OPORDs). We structure our framework by decomposing scenario generation into a hierarchy of subproblems, and for each one, defining the role of the AI tool: (1) generating options for a human author to select from, (2) producing a candidate product for human approval or modification, or (3) generating textual artifacts fully automatically. Our framework employs specialized LLM-based agents to address distinct subproblems. Each agent receives input from preceding subproblem agents, integrating both text-based scenario details and visual information (e.g., map features, unit positions and applies specialized reasoning to produce appropriate outputs. Subsequent agents process these outputs sequentially, preserving logical consistency and ensuring accurate document generation. This multi-agent strategy overcomes the limitations of basic prompting or single-agent approaches when tackling such highly complex tasks. We validate our framework through a proof-of-concept that generates the scheme of maneuver and movement section of an OPORD while estimating map positions and movements as a precursor demonstrating its feasibility and accuracy. Our results demonstrate the potential of LLM-driven multi-agent systems to generate coherent, nuanced documents and adapt dynamically to changing conditions, advancing automation in scenario generation for military training.},
note = {arXiv:2511.07690 [cs.AI]},
keywords = {AI, DTIC, Learning Sciences, LLM, UARC},
pubstate = {published},
tppubtype = {misc}
}
Achieving expert-level performance in simulation-based training relies on the creation of complex, adaptable scenarios, a traditionally laborious and resource intensive process. Although prior research explored scenario generation for military training, pre-LLM AI tools struggled to generate sufficiently complex or adaptable scenarios. This paper introduces a multi-agent, multi-modal reasoning framework that leverages Large Language Models (LLMs) to generate critical training artifacts, such as Operations Orders (OPORDs). We structure our framework by decomposing scenario generation into a hierarchy of subproblems, and for each one, defining the role of the AI tool: (1) generating options for a human author to select from, (2) producing a candidate product for human approval or modification, or (3) generating textual artifacts fully automatically. Our framework employs specialized LLM-based agents to address distinct subproblems. Each agent receives input from preceding subproblem agents, integrating both text-based scenario details and visual information (e.g., map features, unit positions and applies specialized reasoning to produce appropriate outputs. Subsequent agents process these outputs sequentially, preserving logical consistency and ensuring accurate document generation. This multi-agent strategy overcomes the limitations of basic prompting or single-agent approaches when tackling such highly complex tasks. We validate our framework through a proof-of-concept that generates the scheme of maneuver and movement section of an OPORD while estimating map positions and movements as a precursor demonstrating its feasibility and accuracy. Our results demonstrate the potential of LLM-driven multi-agent systems to generate coherent, nuanced documents and adapt dynamically to changing conditions, advancing automation in scenario generation for military training.
Kwon, Deuksin; Shrestha, Kaleen; Han, Bin; Lee, Elena Hayoung; Lucas, Gale
Evaluating Behavioral Alignment in Conflict Dialogue: A Multi-Dimensional Comparison of LLM Agents and Humans Proceedings Article
In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pp. 16377–16391, Association for Computational Linguistics, Suzhou, China, 2025.
Links | BibTeX | Tags: AI, DTIC, LLM, UARC
@inproceedings{kwon_evaluating_2025-1,
title = {Evaluating Behavioral Alignment in Conflict Dialogue: A Multi-Dimensional Comparison of LLM Agents and Humans},
author = {Deuksin Kwon and Kaleen Shrestha and Bin Han and Elena Hayoung Lee and Gale Lucas},
url = {https://aclanthology.org/2025.emnlp-main.828},
doi = {10.18653/v1/2025.emnlp-main.828},
year = {2025},
date = {2025-11-01},
urldate = {2026-06-12},
booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
pages = {16377–16391},
publisher = {Association for Computational Linguistics},
address = {Suzhou, China},
keywords = {AI, DTIC, LLM, UARC},
pubstate = {published},
tppubtype = {inproceedings}
}
Bai, Yuyin; Luo, Yange; Goh, Crystal C W; Rui, Cuiziyi; Gao, Rui; Wu, Yingying; Jiang, Zhongmei; Lu, Lifeng; Rizzo, Albert “Skip”; Bi, Bo
In: J Med Internet Res, vol. 27, pp. e76673–e76673, 2025, ISSN: 1438-8871.
Abstract | Links | BibTeX | Tags: VR
@article{bai_relationships_2025-1,
title = {Relationships Between Parent Ratings of Attention-Deficit/Hyperactivity Disorder Behaviors and the Virtual Reality Attention Tracker in School-Aged Children: Cross-Sectional Study},
author = {Yuyin Bai and Yange Luo and Crystal C W Goh and Cuiziyi Rui and Rui Gao and Yingying Wu and Zhongmei Jiang and Lifeng Lu and Albert “Skip” Rizzo and Bo Bi},
url = {https://www.jmir.org/2025/1/e76673},
doi = {10.2196/76673},
issn = {1438-8871},
year = {2025},
date = {2025-10-01},
urldate = {2026-07-06},
journal = {J Med Internet Res},
volume = {27},
pages = {e76673–e76673},
abstract = {Abstract
Background
Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder with symptoms of inattention, hyperactivity, and impulsivity, which can significantly impact social functioning. Traditional ADHD diagnostic methods rely on subjective behavioral ratings and neuropsychological tests, but these tools may have limitations, including biases and restricted ecological validity.
Objective
This study aimed to explore the relationship between parental ratings of ADHD symptoms and performance on a Virtual Reality Attention Task (VRAT) in school-aged children. The objective was to examine whether the VRAT could provide an objective, ecologically valid measure of ADHD symptoms, and how it correlates with established ADHD rating scales, namely the Swanson Nolan and Pelham, Version IV Scale-Parent Scales (SNAP-IV) and the Chinese version of the Conners’ Parent Rating Scale–Revised (CPRS-48).
Methods
A total of 425 school-aged children (6 to 8 years old) participated in this cross-sectional study. They completed the VRAT, a continuous performance test embedded within an immersive virtual classroom, while their parents completed the SNAP-IV and CPRS-48 rating scales. Bivariate correlational analysis and receiver operating characteristic curve analysis were used to examine the relationships between the VRAT and rating scales, as well as the discriminatory capacity of the VRAT.
Results
VRAT shows several correlations with the SNAP-IV subscales and the CPRS-48 subscales. Participants with higher SNAP-IV inattention and hyperactivity scores exhibited lower VRAT attention and motion performance index scores (all
P
<.001). For the CPRS-48, significant correlations were noted with conduct problems, learning problems, psychosomatic problems, impulsivity-hyperactivity, and attention index, particularly with the attention performance and motion performance indexes (all
P
<.05). Gender differences were significant in attention and hyperactivity measures of the SNAP-IV and CPRS-48, while only latency showed gender differences in the VRAT. In addition, the receiver operating characteristic analysis indicated that the key performance metrics of VRAT showed moderate discriminatory power, with area under the curve values varying from 0.56 to 0.74.
Conclusions
This study highlights the potential of virtual reality–based assessments such as the VRAT in ADHD diagnostics, providing an innovative approach to evaluating attention in a more immersive and ecologically valid setting. However, given the modest correlation with parent rating scales, it suggests that a combination of objective and subjective assessment tools would provide the most accurate and comprehensive ADHD diagnosis.},
keywords = {VR},
pubstate = {published},
tppubtype = {article}
}
Abstract
Background
Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder with symptoms of inattention, hyperactivity, and impulsivity, which can significantly impact social functioning. Traditional ADHD diagnostic methods rely on subjective behavioral ratings and neuropsychological tests, but these tools may have limitations, including biases and restricted ecological validity.
Objective
This study aimed to explore the relationship between parental ratings of ADHD symptoms and performance on a Virtual Reality Attention Task (VRAT) in school-aged children. The objective was to examine whether the VRAT could provide an objective, ecologically valid measure of ADHD symptoms, and how it correlates with established ADHD rating scales, namely the Swanson Nolan and Pelham, Version IV Scale-Parent Scales (SNAP-IV) and the Chinese version of the Conners’ Parent Rating Scale–Revised (CPRS-48).
Methods
A total of 425 school-aged children (6 to 8 years old) participated in this cross-sectional study. They completed the VRAT, a continuous performance test embedded within an immersive virtual classroom, while their parents completed the SNAP-IV and CPRS-48 rating scales. Bivariate correlational analysis and receiver operating characteristic curve analysis were used to examine the relationships between the VRAT and rating scales, as well as the discriminatory capacity of the VRAT.
Results
VRAT shows several correlations with the SNAP-IV subscales and the CPRS-48 subscales. Participants with higher SNAP-IV inattention and hyperactivity scores exhibited lower VRAT attention and motion performance index scores (all
P
<.001). For the CPRS-48, significant correlations were noted with conduct problems, learning problems, psychosomatic problems, impulsivity-hyperactivity, and attention index, particularly with the attention performance and motion performance indexes (all
P
<.05). Gender differences were significant in attention and hyperactivity measures of the SNAP-IV and CPRS-48, while only latency showed gender differences in the VRAT. In addition, the receiver operating characteristic analysis indicated that the key performance metrics of VRAT showed moderate discriminatory power, with area under the curve values varying from 0.56 to 0.74.
Conclusions
This study highlights the potential of virtual reality–based assessments such as the VRAT in ADHD diagnostics, providing an innovative approach to evaluating attention in a more immersive and ecologically valid setting. However, given the modest correlation with parent rating scales, it suggests that a combination of objective and subjective assessment tools would provide the most accurate and comprehensive ADHD diagnosis.
Background
Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder with symptoms of inattention, hyperactivity, and impulsivity, which can significantly impact social functioning. Traditional ADHD diagnostic methods rely on subjective behavioral ratings and neuropsychological tests, but these tools may have limitations, including biases and restricted ecological validity.
Objective
This study aimed to explore the relationship between parental ratings of ADHD symptoms and performance on a Virtual Reality Attention Task (VRAT) in school-aged children. The objective was to examine whether the VRAT could provide an objective, ecologically valid measure of ADHD symptoms, and how it correlates with established ADHD rating scales, namely the Swanson Nolan and Pelham, Version IV Scale-Parent Scales (SNAP-IV) and the Chinese version of the Conners’ Parent Rating Scale–Revised (CPRS-48).
Methods
A total of 425 school-aged children (6 to 8 years old) participated in this cross-sectional study. They completed the VRAT, a continuous performance test embedded within an immersive virtual classroom, while their parents completed the SNAP-IV and CPRS-48 rating scales. Bivariate correlational analysis and receiver operating characteristic curve analysis were used to examine the relationships between the VRAT and rating scales, as well as the discriminatory capacity of the VRAT.
Results
VRAT shows several correlations with the SNAP-IV subscales and the CPRS-48 subscales. Participants with higher SNAP-IV inattention and hyperactivity scores exhibited lower VRAT attention and motion performance index scores (all
P
<.001). For the CPRS-48, significant correlations were noted with conduct problems, learning problems, psychosomatic problems, impulsivity-hyperactivity, and attention index, particularly with the attention performance and motion performance indexes (all
P
<.05). Gender differences were significant in attention and hyperactivity measures of the SNAP-IV and CPRS-48, while only latency showed gender differences in the VRAT. In addition, the receiver operating characteristic analysis indicated that the key performance metrics of VRAT showed moderate discriminatory power, with area under the curve values varying from 0.56 to 0.74.
Conclusions
This study highlights the potential of virtual reality–based assessments such as the VRAT in ADHD diagnostics, providing an innovative approach to evaluating attention in a more immersive and ecologically valid setting. However, given the modest correlation with parent rating scales, it suggests that a combination of objective and subjective assessment tools would provide the most accurate and comprehensive ADHD diagnosis.
Zhang, Tianyi; Zhou, Xiaolin; Wang, Yunzhe; Cambria, Erik; Traum, David; Mao, Rui
Individualized Cognitive Simulation in Large Language Models: Evaluating Different Cognitive Representation Methods Miscellaneous
2025, (arXiv:2510.20252 [cs.AI]).
Abstract | Links | BibTeX | Tags: AI, LLM, Natural Language, Simulation
@misc{zhang_individualized_2025,
title = {Individualized Cognitive Simulation in Large Language Models: Evaluating Different Cognitive Representation Methods},
author = {Tianyi Zhang and Xiaolin Zhou and Yunzhe Wang and Erik Cambria and David Traum and Rui Mao},
url = {http://arxiv.org/abs/2510.20252},
doi = {10.48550/arXiv.2510.20252},
year = {2025},
date = {2025-10-01},
urldate = {2026-07-06},
publisher = {arXiv},
abstract = {Individualized cognitive simulation (ICS) aims to build computational models that approximate the thought processes of specific individuals. While large language models (LLMs) convincingly mimic surface-level human behavior such as role-play, their ability to simulate deeper individualized cognitive processes remains poorly understood. To address this gap, we introduce a novel task that evaluates different cognitive representation methods in ICS. We construct a dataset from recently published novels (later than the release date of the tested LLMs) and propose an 11-condition cognitive evaluation framework to benchmark seven off-the-shelf LLMs in the context of authorial style emulation. We hypothesize that effective cognitive representations can help LLMs generate storytelling that better mirrors the original author. Thus, we test different cognitive representations, e.g., linguistic features, concept mappings, and profile-based information. Results show that combining conceptual and linguistic features is particularly effective in ICS, outperforming static profile-based cues in overall evaluation. Importantly, LLMs are more effective at mimicking linguistic style than narrative structure, underscoring their limits in deeper cognitive simulation. These findings provide a foundation for developing AI systems that adapt to individual ways of thinking and expression, advancing more personalized and human-aligned creative technologies.},
note = {arXiv:2510.20252 [cs.AI]},
keywords = {AI, LLM, Natural Language, Simulation},
pubstate = {published},
tppubtype = {misc}
}
Individualized cognitive simulation (ICS) aims to build computational models that approximate the thought processes of specific individuals. While large language models (LLMs) convincingly mimic surface-level human behavior such as role-play, their ability to simulate deeper individualized cognitive processes remains poorly understood. To address this gap, we introduce a novel task that evaluates different cognitive representation methods in ICS. We construct a dataset from recently published novels (later than the release date of the tested LLMs) and propose an 11-condition cognitive evaluation framework to benchmark seven off-the-shelf LLMs in the context of authorial style emulation. We hypothesize that effective cognitive representations can help LLMs generate storytelling that better mirrors the original author. Thus, we test different cognitive representations, e.g., linguistic features, concept mappings, and profile-based information. Results show that combining conceptual and linguistic features is particularly effective in ICS, outperforming static profile-based cues in overall evaluation. Importantly, LLMs are more effective at mimicking linguistic style than narrative structure, underscoring their limits in deeper cognitive simulation. These findings provide a foundation for developing AI systems that adapt to individual ways of thinking and expression, advancing more personalized and human-aligned creative technologies.
Bai, Yuyin; Luo, Yange; Goh, Crystal C. W.; Rui, Cuiziyi; Gao, Rui; Wu, Yingying; Jiang, Zhongmei; Lu, Lifeng; Rizzo, Albert Skip; Bi, Bo
In: J Med Internet Res, vol. 27, pp. e76673, 2025, ISSN: 1438-8871.
Abstract | Links | BibTeX | Tags: MedVR, VR
@article{bai_relationships_2025,
title = {Relationships Between Parent Ratings of Attention-Deficit/Hyperactivity Disorder Behaviors and the Virtual Reality Attention Tracker in School-Aged Children: Cross-Sectional Study},
author = {Yuyin Bai and Yange Luo and Crystal C. W. Goh and Cuiziyi Rui and Rui Gao and Yingying Wu and Zhongmei Jiang and Lifeng Lu and Albert Skip Rizzo and Bo Bi},
doi = {10.2196/76673},
issn = {1438-8871},
year = {2025},
date = {2025-10-01},
journal = {J Med Internet Res},
volume = {27},
pages = {e76673},
abstract = {BACKGROUND: Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder with symptoms of inattention, hyperactivity, and impulsivity, which can significantly impact social functioning. Traditional ADHD diagnostic methods rely on subjective behavioral ratings and neuropsychological tests, but these tools may have limitations, including biases and restricted ecological validity.
OBJECTIVE: This study aimed to explore the relationship between parental ratings of ADHD symptoms and performance on a Virtual Reality Attention Task (VRAT) in school-aged children. The objective was to examine whether the VRAT could provide an objective, ecologically valid measure of ADHD symptoms, and how it correlates with established ADHD rating scales, namely the Swanson Nolan and Pelham, Version IV Scale-Parent Scales (SNAP-IV) and the Chinese version of the Conners' Parent Rating Scale-Revised (CPRS-48).
METHODS: A total of 425 school-aged children (6 to 8 years old) participated in this cross-sectional study. They completed the VRAT, a continuous performance test embedded within an immersive virtual classroom, while their parents completed the SNAP-IV and CPRS-48 rating scales. Bivariate correlational analysis and receiver operating characteristic curve analysis were used to examine the relationships between the VRAT and rating scales, as well as the discriminatory capacity of the VRAT.},
keywords = {MedVR, VR},
pubstate = {published},
tppubtype = {article}
}
BACKGROUND: Attention-deficit/hyperactivity disorder (ADHD) is a common neurodevelopmental disorder with symptoms of inattention, hyperactivity, and impulsivity, which can significantly impact social functioning. Traditional ADHD diagnostic methods rely on subjective behavioral ratings and neuropsychological tests, but these tools may have limitations, including biases and restricted ecological validity.
OBJECTIVE: This study aimed to explore the relationship between parental ratings of ADHD symptoms and performance on a Virtual Reality Attention Task (VRAT) in school-aged children. The objective was to examine whether the VRAT could provide an objective, ecologically valid measure of ADHD symptoms, and how it correlates with established ADHD rating scales, namely the Swanson Nolan and Pelham, Version IV Scale-Parent Scales (SNAP-IV) and the Chinese version of the Conners' Parent Rating Scale-Revised (CPRS-48).
METHODS: A total of 425 school-aged children (6 to 8 years old) participated in this cross-sectional study. They completed the VRAT, a continuous performance test embedded within an immersive virtual classroom, while their parents completed the SNAP-IV and CPRS-48 rating scales. Bivariate correlational analysis and receiver operating characteristic curve analysis were used to examine the relationships between the VRAT and rating scales, as well as the discriminatory capacity of the VRAT.
OBJECTIVE: This study aimed to explore the relationship between parental ratings of ADHD symptoms and performance on a Virtual Reality Attention Task (VRAT) in school-aged children. The objective was to examine whether the VRAT could provide an objective, ecologically valid measure of ADHD symptoms, and how it correlates with established ADHD rating scales, namely the Swanson Nolan and Pelham, Version IV Scale-Parent Scales (SNAP-IV) and the Chinese version of the Conners' Parent Rating Scale-Revised (CPRS-48).
METHODS: A total of 425 school-aged children (6 to 8 years old) participated in this cross-sectional study. They completed the VRAT, a continuous performance test embedded within an immersive virtual classroom, while their parents completed the SNAP-IV and CPRS-48 rating scales. Bivariate correlational analysis and receiver operating characteristic curve analysis were used to examine the relationships between the VRAT and rating scales, as well as the discriminatory capacity of the VRAT.
Gordon, Andrew S.
Automated Formalization of William Shakespeare’s Venus and Adonis Using Large Language Models Proceedings Article
In: 12th Annual Conference on Advances in Cognitive Systems, pp. 259–274, Atlanta, GA, 2025.
Abstract | Links | BibTeX | Tags: ARO, Cognitive Architecture, DTIC, LLM, UARC
@inproceedings{gordon_automated_2025,
title = {Automated Formalization of William Shakespeare’s Venus and Adonis Using Large Language Models},
author = {Andrew S. Gordon},
url = {https://asgordon.github.io/publications/ACS2025.PDF},
year = {2025},
date = {2025-10-01},
booktitle = {12th Annual Conference on Advances in Cognitive Systems},
volume = {ACS-2025 (259-274)},
pages = {259–274},
address = {Atlanta, GA},
abstract = {Advances in large language models have created new opportunities for the translation of natural
language into formal representations for use in symbolic reasoning systems, particularly when the
formal vocabulary is well-specified with sufficient training examples. In this paper, we explore the
more challenging problem of automating the knowledge engineering task of authoring the formal
vocabulary itself, based on the representational needs of the input text. We describe a methodology
that incrementally builds a formal vocabulary of first-order predicate-argument definitions and constant terms by prompting a large language model and restricting its output generations to adhere to
tightly constrained formalisms. We evaluate this methodology in its application to a challenging literary text, William Shakespeare’s narrative poem “Venus and Adonis” (1593), with a focus on how
well new vocabulary is reused by the large language model in after it has been introduced. We then
investigate the semantic content of the generated formalisms by manually sorting the vocabulary
into a taxonomy of 48 foundational areas used in knowledge representation research. Our results
point to several challenges in fully-automated knowledge engineering pipelines, but also point to
new opportunities in using large language models to support axiomization of foundational theories.},
keywords = {ARO, Cognitive Architecture, DTIC, LLM, UARC},
pubstate = {published},
tppubtype = {inproceedings}
}
Advances in large language models have created new opportunities for the translation of natural
language into formal representations for use in symbolic reasoning systems, particularly when the
formal vocabulary is well-specified with sufficient training examples. In this paper, we explore the
more challenging problem of automating the knowledge engineering task of authoring the formal
vocabulary itself, based on the representational needs of the input text. We describe a methodology
that incrementally builds a formal vocabulary of first-order predicate-argument definitions and constant terms by prompting a large language model and restricting its output generations to adhere to
tightly constrained formalisms. We evaluate this methodology in its application to a challenging literary text, William Shakespeare’s narrative poem “Venus and Adonis” (1593), with a focus on how
well new vocabulary is reused by the large language model in after it has been introduced. We then
investigate the semantic content of the generated formalisms by manually sorting the vocabulary
into a taxonomy of 48 foundational areas used in knowledge representation research. Our results
point to several challenges in fully-automated knowledge engineering pipelines, but also point to
new opportunities in using large language models to support axiomization of foundational theories.
language into formal representations for use in symbolic reasoning systems, particularly when the
formal vocabulary is well-specified with sufficient training examples. In this paper, we explore the
more challenging problem of automating the knowledge engineering task of authoring the formal
vocabulary itself, based on the representational needs of the input text. We describe a methodology
that incrementally builds a formal vocabulary of first-order predicate-argument definitions and constant terms by prompting a large language model and restricting its output generations to adhere to
tightly constrained formalisms. We evaluate this methodology in its application to a challenging literary text, William Shakespeare’s narrative poem “Venus and Adonis” (1593), with a focus on how
well new vocabulary is reused by the large language model in after it has been introduced. We then
investigate the semantic content of the generated formalisms by manually sorting the vocabulary
into a taxonomy of 48 foundational areas used in knowledge representation research. Our results
point to several challenges in fully-automated knowledge engineering pipelines, but also point to
new opportunities in using large language models to support axiomization of foundational theories.
Wang, Yunzhe; Hans, Soham; Ustun, Volkan
X-Ego: Acquiring Team-Level Tactical Situational Awareness via Cross-Egocentric Contrastive Video Representation Learning Miscellaneous
2025, (arXiv:2510.19150 [cs.CV]).
Abstract | Links | BibTeX | Tags: Cognitive Architecture, DTIC, UARC
@misc{wang_x-ego_2025,
title = {X-Ego: Acquiring Team-Level Tactical Situational Awareness via Cross-Egocentric Contrastive Video Representation Learning},
author = {Yunzhe Wang and Soham Hans and Volkan Ustun},
url = {http://arxiv.org/abs/2510.19150},
doi = {10.48550/arXiv.2510.19150},
year = {2025},
date = {2025-10-01},
urldate = {2026-05-29},
publisher = {arXiv},
abstract = {Human team tactics emerge from each player's individual perspective and their ability to anticipate, interpret, and adapt to teammates' intentions. While advances in video understanding have improved the modeling of team interactions in sports, most existing work relies on third-person broadcast views and overlooks the synchronous, egocentric nature of multi-agent learning. We introduce X-Ego-CS, a benchmark dataset consisting of 124 hours of gameplay footage from 45 professional-level matches of the popular e-sports game Counter-Strike 2, designed to facilitate research on multi-agent decision-making in complex 3D environments. X-Ego-CS provides cross-egocentric video streams that synchronously capture all players' first-person perspectives along with state-action trajectories. Building on this resource, we propose Cross-Ego Contrastive Learning (CECL), which aligns teammates' egocentric visual streams to foster team-level tactical situational awareness from an individual's perspective. We evaluate CECL on a teammate-opponent location prediction task, demonstrating its effectiveness in enhancing an agent's ability to infer both teammate and opponent positions from a single first-person view using state-of-the-art video encoders. Together, X-Ego-CS and CECL establish a foundation for cross-egocentric multi-agent benchmarking in esports. More broadly, our work positions gameplay understanding as a testbed for multi-agent modeling and tactical learning, with implications for spatiotemporal reasoning and human-AI teaming in both virtual and real-world domains. Code and dataset are available at https://github.com/HATS-ICT/x-ego.},
note = {arXiv:2510.19150 [cs.CV]},
keywords = {Cognitive Architecture, DTIC, UARC},
pubstate = {published},
tppubtype = {misc}
}
Human team tactics emerge from each player's individual perspective and their ability to anticipate, interpret, and adapt to teammates' intentions. While advances in video understanding have improved the modeling of team interactions in sports, most existing work relies on third-person broadcast views and overlooks the synchronous, egocentric nature of multi-agent learning. We introduce X-Ego-CS, a benchmark dataset consisting of 124 hours of gameplay footage from 45 professional-level matches of the popular e-sports game Counter-Strike 2, designed to facilitate research on multi-agent decision-making in complex 3D environments. X-Ego-CS provides cross-egocentric video streams that synchronously capture all players' first-person perspectives along with state-action trajectories. Building on this resource, we propose Cross-Ego Contrastive Learning (CECL), which aligns teammates' egocentric visual streams to foster team-level tactical situational awareness from an individual's perspective. We evaluate CECL on a teammate-opponent location prediction task, demonstrating its effectiveness in enhancing an agent's ability to infer both teammate and opponent positions from a single first-person view using state-of-the-art video encoders. Together, X-Ego-CS and CECL establish a foundation for cross-egocentric multi-agent benchmarking in esports. More broadly, our work positions gameplay understanding as a testbed for multi-agent modeling and tactical learning, with implications for spatiotemporal reasoning and human-AI teaming in both virtual and real-world domains. Code and dataset are available at https://github.com/HATS-ICT/x-ego.
Han, Bin; Hoegen, Jessie; Lei, Su; Lucas, Gale; Shore, Danielle; Parkinson, Brian; Gratch, Jonathan
How Expression, Context and Perspective Determine Judgments of Emotion Journal Article
In: IEEE Trans. Affective Comput., pp. 1–12, 2025, ISSN: 1949-3045, 2371-9850.
Links | BibTeX | Tags: DTIC, Emotion
@article{han_how_2025,
title = {How Expression, Context and Perspective Determine Judgments of Emotion},
author = {Bin Han and Jessie Hoegen and Su Lei and Gale Lucas and Danielle Shore and Brian Parkinson and Jonathan Gratch},
url = {https://ieeexplore.ieee.org/document/11194717/},
doi = {10.1109/TAFFC.2025.3618588},
issn = {1949-3045, 2371-9850},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-16},
journal = {IEEE Trans. Affective Comput.},
pages = {1–12},
keywords = {DTIC, Emotion},
pubstate = {published},
tppubtype = {article}
}
Banayeeanzade, Amin; Tak, Ala N.; Bahrani, Fatemeh; Bolourani, Anahita; Blas, Leonardo; Ferrara, Emilio; Gratch, Jonathan; Karimireddy, Sai Praneeth
Psychological Steering in LLMs: An Evaluation of Effectiveness and Trustworthiness Miscellaneous
2025, (arXiv:2510.04484 [cs]).
Abstract | Links | BibTeX | Tags: AI
@misc{banayeeanzade_psychological_2025,
title = {Psychological Steering in LLMs: An Evaluation of Effectiveness and Trustworthiness},
author = {Amin Banayeeanzade and Ala N. Tak and Fatemeh Bahrani and Anahita Bolourani and Leonardo Blas and Emilio Ferrara and Jonathan Gratch and Sai Praneeth Karimireddy},
url = {http://arxiv.org/abs/2510.04484},
doi = {10.48550/arXiv.2510.04484},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-16},
publisher = {arXiv},
abstract = {The ability to control LLMs' emulated emotional states and personality traits is essential for enabling rich, human-centered interactions in socially interactive settings. We introduce PsySET, a Psychologically-informed benchmark to evaluate LLM Steering Effectiveness and Trustworthiness across the emotion and personality domains. Our study spans four models from different LLM families paired with various steering strategies, including prompting, fine-tuning, and representation engineering. Our results indicate that prompting is consistently effective but limited in intensity control, whereas vector injections achieve finer controllability while slightly reducing output quality. Moreover, we explore the trustworthiness of steered LLMs by assessing safety, truthfulness, fairness, and ethics, highlighting potential side effects and behavioral shifts. Notably, we observe idiosyncratic effects; for instance, even a positive emotion like joy can degrade robustness to adversarial factuality, lower privacy awareness, and increase preferential bias. Meanwhile, anger predictably elevates toxicity yet strengthens leakage resistance. Our framework establishes the first holistic evaluation of emotion and personality steering, offering insights into its interpretability and reliability for socially interactive applications.},
note = {arXiv:2510.04484 [cs]},
keywords = {AI},
pubstate = {published},
tppubtype = {misc}
}
The ability to control LLMs' emulated emotional states and personality traits is essential for enabling rich, human-centered interactions in socially interactive settings. We introduce PsySET, a Psychologically-informed benchmark to evaluate LLM Steering Effectiveness and Trustworthiness across the emotion and personality domains. Our study spans four models from different LLM families paired with various steering strategies, including prompting, fine-tuning, and representation engineering. Our results indicate that prompting is consistently effective but limited in intensity control, whereas vector injections achieve finer controllability while slightly reducing output quality. Moreover, we explore the trustworthiness of steered LLMs by assessing safety, truthfulness, fairness, and ethics, highlighting potential side effects and behavioral shifts. Notably, we observe idiosyncratic effects; for instance, even a positive emotion like joy can degrade robustness to adversarial factuality, lower privacy awareness, and increase preferential bias. Meanwhile, anger predictably elevates toxicity yet strengthens leakage resistance. Our framework establishes the first holistic evaluation of emotion and personality steering, offering insights into its interpretability and reliability for socially interactive applications.
Rizk, Basem; Walsh, Joel; Core, Mark; Nye, Benjamin
From Videos to Indexed Knowledge Graphs – Framework to Marry Methods for Multimodal Content Analysis and Understanding Miscellaneous
2025, (arXiv:2510.01513 [cs]).
Abstract | Links | BibTeX | Tags: DTIC, Narrative
@misc{rizk_videos_2025,
title = {From Videos to Indexed Knowledge Graphs – Framework to Marry Methods for Multimodal Content Analysis and Understanding},
author = {Basem Rizk and Joel Walsh and Mark Core and Benjamin Nye},
url = {http://arxiv.org/abs/2510.01513},
doi = {10.48550/arXiv.2510.01513},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-16},
publisher = {arXiv},
abstract = {Analysis of multi-modal content can be tricky, computationally expensive, and require a significant amount of engineering efforts. Lots of work with pre-trained models on static data is out there, yet fusing these opensource models and methods with complex data such as videos is relatively challenging. In this paper, we present a framework that enables efficiently prototyping pipelines for multi-modal content analysis. We craft a candidate recipe for a pipeline, marrying a set of pre-trained models, to convert videos into a temporal semi-structured data format. We translate this structure further to a frame-level indexed knowledge graph representation that is query-able and supports continual learning, enabling the dynamic incorporation of new domain-specific knowledge through an interactive medium.},
note = {arXiv:2510.01513 [cs]},
keywords = {DTIC, Narrative},
pubstate = {published},
tppubtype = {misc}
}
Analysis of multi-modal content can be tricky, computationally expensive, and require a significant amount of engineering efforts. Lots of work with pre-trained models on static data is out there, yet fusing these opensource models and methods with complex data such as videos is relatively challenging. In this paper, we present a framework that enables efficiently prototyping pipelines for multi-modal content analysis. We craft a candidate recipe for a pipeline, marrying a set of pre-trained models, to convert videos into a temporal semi-structured data format. We translate this structure further to a frame-level indexed knowledge graph representation that is query-able and supports continual learning, enabling the dynamic incorporation of new domain-specific knowledge through an interactive medium.
Tran, Minh; Siniukov, Maksim; Jin, Zhangyu; Soleymani, Mohammad
Discrete Facial Encoding: : A Framework for Data-driven Facial Display Discovery Miscellaneous
2025, (arXiv:2510.01662 [cs]).
Abstract | Links | BibTeX | Tags:
@misc{tran_discrete_2025,
title = {Discrete Facial Encoding: : A Framework for Data-driven Facial Display Discovery},
author = {Minh Tran and Maksim Siniukov and Zhangyu Jin and Mohammad Soleymani},
url = {http://arxiv.org/abs/2510.01662},
doi = {10.48550/arXiv.2510.01662},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-16},
publisher = {arXiv},
abstract = {Facial expression analysis is central to understanding human behavior, yet existing coding systems such as the Facial Action Coding System (FACS) are constrained by limited coverage and costly manual annotation. In this work, we introduce Discrete Facial Encoding (DFE), an unsupervised, data-driven alternative of compact and interpretable dictionary of facial expressions from 3D mesh sequences learned through a Residual Vector Quantized Variational Autoencoder (RVQ-VAE). Our approach first extracts identity-invariant expression features from images using a 3D Morphable Model (3DMM), effectively disentangling factors such as head pose and facial geometry. We then encode these features using an RVQ-VAE, producing a sequence of discrete tokens from a shared codebook, where each token captures a specific, reusable facial deformation pattern that contributes to the overall expression. Through extensive experiments, we demonstrate that Discrete Facial Encoding captures more precise facial behaviors than FACS and other facial encoding alternatives. We evaluate the utility of our representation across three high-level psychological tasks: stress detection, personality prediction, and depression detection. Using a simple Bag-of-Words model built on top of the learned tokens, our system consistently outperforms both FACS-based pipelines and strong image and video representation learning models such as Masked Autoencoders. Further analysis reveals that our representation covers a wider variety of facial displays, highlighting its potential as a scalable and effective alternative to FACS for psychological and affective computing applications.},
note = {arXiv:2510.01662 [cs]},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Facial expression analysis is central to understanding human behavior, yet existing coding systems such as the Facial Action Coding System (FACS) are constrained by limited coverage and costly manual annotation. In this work, we introduce Discrete Facial Encoding (DFE), an unsupervised, data-driven alternative of compact and interpretable dictionary of facial expressions from 3D mesh sequences learned through a Residual Vector Quantized Variational Autoencoder (RVQ-VAE). Our approach first extracts identity-invariant expression features from images using a 3D Morphable Model (3DMM), effectively disentangling factors such as head pose and facial geometry. We then encode these features using an RVQ-VAE, producing a sequence of discrete tokens from a shared codebook, where each token captures a specific, reusable facial deformation pattern that contributes to the overall expression. Through extensive experiments, we demonstrate that Discrete Facial Encoding captures more precise facial behaviors than FACS and other facial encoding alternatives. We evaluate the utility of our representation across three high-level psychological tasks: stress detection, personality prediction, and depression detection. Using a simple Bag-of-Words model built on top of the learned tokens, our system consistently outperforms both FACS-based pipelines and strong image and video representation learning models such as Masked Autoencoders. Further analysis reveals that our representation covers a wider variety of facial displays, highlighting its potential as a scalable and effective alternative to FACS for psychological and affective computing applications.
Charnsethikul, Pithayuth; Zunquti, Almajd; Lucas, Gale; Mirkovic, Jelena
Navigating Social Media Privacy: Awareness, Preferences, and Discoverability Journal Article
In: PoPETs, vol. 2025, no. 4, pp. 620–638, 2025, ISSN: 2299-0984.
Abstract | Links | BibTeX | Tags: DTIC, Social
@article{charnsethikul_navigating_2025,
title = {Navigating Social Media Privacy: Awareness, Preferences, and Discoverability},
author = {Pithayuth Charnsethikul and Almajd Zunquti and Gale Lucas and Jelena Mirkovic},
url = {https://petsymposium.org/popets/2025/popets-2025-0148.php},
doi = {10.56553/popets-2025-0148},
issn = {2299-0984},
year = {2025},
date = {2025-10-01},
urldate = {2025-08-19},
journal = {PoPETs},
volume = {2025},
number = {4},
pages = {620–638},
abstract = {Social media platforms provide various privacy settings, which users can adjust to fit their privacy needs. Platforms claim that this is sufficient – users have power to accept the default settings they like, and change those they do not like. In this paper, we seek to quantify user awareness of, preferences around and ability to adjust social media privacy settings. We conduct an online survey of 541 participants across six different social media platforms: Facebook, Instagram, X, LinkedIn, TikTok, and Snapchat. We focus on nine privacy settings that are commonly available across these platforms, and evaluate participants’ preferences for privacy, awareness of the privacy settings and ability to locate them. We find that default settings are ill-aligned with user preferences – 92% of participants prefer at least one of the privacy options to be more private than the default. We further find that users are generally not aware of privacy settings, and struggle to find them. 80% of participants have never seen at least one privacy setting, and 79% of participants rated at least one setting as hard to find. We also find that the fewer privacy settings a user has seen, the harder for them to locate those settings, and the higher the level of privacy they desire. Additionally, we find that there are significant differences in privacy setting preferences and usability across different user age groups and across platforms. Older users are more conservative about their privacy, they have seen significantly fewer privacy settings, and they spend significantly more time locating them than younger users. On some platforms, like LinkedIn, users opt for higher visibility, while on others they prefer more privacy. Some platforms, like TikTok, make it significantly easier for users to locate privacy settings. Based on our findings, we provide recommendations on default values and how to improve usability of privacy settings on social media.},
keywords = {DTIC, Social},
pubstate = {published},
tppubtype = {article}
}
Social media platforms provide various privacy settings, which users can adjust to fit their privacy needs. Platforms claim that this is sufficient – users have power to accept the default settings they like, and change those they do not like. In this paper, we seek to quantify user awareness of, preferences around and ability to adjust social media privacy settings. We conduct an online survey of 541 participants across six different social media platforms: Facebook, Instagram, X, LinkedIn, TikTok, and Snapchat. We focus on nine privacy settings that are commonly available across these platforms, and evaluate participants’ preferences for privacy, awareness of the privacy settings and ability to locate them. We find that default settings are ill-aligned with user preferences – 92% of participants prefer at least one of the privacy options to be more private than the default. We further find that users are generally not aware of privacy settings, and struggle to find them. 80% of participants have never seen at least one privacy setting, and 79% of participants rated at least one setting as hard to find. We also find that the fewer privacy settings a user has seen, the harder for them to locate those settings, and the higher the level of privacy they desire. Additionally, we find that there are significant differences in privacy setting preferences and usability across different user age groups and across platforms. Older users are more conservative about their privacy, they have seen significantly fewer privacy settings, and they spend significantly more time locating them than younger users. On some platforms, like LinkedIn, users opt for higher visibility, while on others they prefer more privacy. Some platforms, like TikTok, make it significantly easier for users to locate privacy settings. Based on our findings, we provide recommendations on default values and how to improve usability of privacy settings on social media.
Wang, Yunzhe; Ustun, Volkan; McGroarty, Chris
A Data-Driven Discretized CS:GO Simulation Environment to Facilitate Strategic Multi-Agent Planning Research Miscellaneous
2025, (arXiv:2509.06355 [cs.AI]).
Abstract | Links | BibTeX | Tags: DTIC, Simulation, UARC
@misc{wang_data-driven_2025-1,
title = {A Data-Driven Discretized CS:GO Simulation Environment to Facilitate Strategic Multi-Agent Planning Research},
author = {Yunzhe Wang and Volkan Ustun and Chris McGroarty},
url = {http://arxiv.org/abs/2509.06355},
doi = {10.48550/arXiv.2509.06355},
year = {2025},
date = {2025-09-01},
urldate = {2026-06-22},
publisher = {arXiv},
abstract = {Modern simulation environments for complex multi-agent interactions must balance high-fidelity detail with computational efficiency. We present DECOY, a novel multi-agent simulator that abstracts strategic, long-horizon planning in 3D terrains into high-level discretized simulation while preserving low-level environmental fidelity. Using Counter-Strike: Global Offensive (CS:GO) as a testbed, our framework accurately simulates gameplay using only movement decisions as tactical positioning – without explicitly modeling low-level mechanics such as aiming and shooting. Central to our approach is a waypoint system that simplifies and discretizes continuous states and actions, paired with neural predictive and generative models trained on real CS:GO tournament data to reconstruct event outcomes. Extensive evaluations show that replays generated from human data in DECOY closely match those observed in the original game. Our publicly available simulation environment provides a valuable tool for advancing research in strategic multi-agent planning and behavior generation.},
note = {arXiv:2509.06355 [cs.AI]},
keywords = {DTIC, Simulation, UARC},
pubstate = {published},
tppubtype = {misc}
}
Modern simulation environments for complex multi-agent interactions must balance high-fidelity detail with computational efficiency. We present DECOY, a novel multi-agent simulator that abstracts strategic, long-horizon planning in 3D terrains into high-level discretized simulation while preserving low-level environmental fidelity. Using Counter-Strike: Global Offensive (CS:GO) as a testbed, our framework accurately simulates gameplay using only movement decisions as tactical positioning – without explicitly modeling low-level mechanics such as aiming and shooting. Central to our approach is a waypoint system that simplifies and discretizes continuous states and actions, paired with neural predictive and generative models trained on real CS:GO tournament data to reconstruct event outcomes. Extensive evaluations show that replays generated from human data in DECOY closely match those observed in the original game. Our publicly available simulation environment provides a valuable tool for advancing research in strategic multi-agent planning and behavior generation.