The USC Institute for Creative Technologies (ICT) will present four research papers at the Interservice/Industry Training, Simulation and Education Conference (I/ITSEC) 2025 in Orlando, Florida this December.
The Interservice/Industry Training, Simulation and Education Conference (I/ITSEC) is the world’s largest and most influential event dedicated to modeling, simulation, and training. Held annually in December in Orlando, Florida, I/ITSEC brings together thousands of professionals from defense, industry, academia, and international allies to explore the technologies, strategies, and partnerships shaping the future of training and operational readiness.
ICT Papers at I/ITSEC 2025
1. Towards AI-Assisted Generation of Military Training Scenarios
Authors: Soham Hans, Volkan Ustun, Mark Core, Benjamin Nye (USC Institute for Creative Technologies); James Sterrett, Matthew Green (U.S. Army University Command and General Staff College)
This paper introduces a multi-agent, multi-modal reasoning framework leveraging Large Language Models (LLMs) to automate the generation of complex military training artifacts, such as Operations Orders (OPORDs). The framework employs specialized LLM-based agents that process both text-based scenario details and visual information to produce coherent, nuanced documents while significantly reducing manual workload and accelerating development timelines.
2. Training Developer Feedback on AI for Revision of Content (ARC)
Authors: Benjamin D. Nye, Jose-Luis Ambite, Joel Mathew, Mark G. Core, Daniel Auerbach, Dilan R. Ramirez, Joel Walsh (University of Southern California)
The ARC tool addresses the challenge of rapidly updating military training materials in response to doctrine changes. Using AI to analyze doctrine and training documents, ARC suggests which sections need updating—a critical capability when a single doctrine change can require updates across many courses and training resources. Beta testing with four Army Centers showed positive results, with participants strongly agreeing that ARC would “increase productivity” (average rating 5.5 on a 6-point scale).
3. TopoGen: Training Generative AI to Produce Maps for Experiential Scenarios
Authors: Sophia Khan, Joel Walsh, & Benjamin Nye (University of Southern California Institute for Creative Technologies)
TopoGen addresses the time-consuming challenge of creating specialized maps for scenario-based military training. This work presents a two-part dataset used to train generative diffusion models that can create high-quality military training maps from simple text prompts describing geographic features. The system can generate maps specifying size, relative locations of topography (hills, mountains), and water features (rivers, coasts), accelerating the development of training scenarios.
4. Quantifying Loss Aversion in Cyber Adversaries via LLM Analysis
Authors: Soham Hans, Nikolos Gurney, Stacy Marsella (USC Institute for Creative Technologies); Sofia Hirschmann (Northeastern University)
This research presents a methodology leveraging large language models to extract quantifiable insights into cognitive biases—specifically loss aversion—from hacker behavior. By analyzing data from experiments where hackers attacked a controlled demonstration network, the research demonstrates how LLMs can effectively dissect nuanced behavioral patterns, offering an approach to enhancing cyber defense strategies through real-time, behavior-based analysis.
Serious Games Showcase Finalist
Becoming Fei has been selected as a finalist for the I/ITSEC Serious Games Showcase and Competition. Developed through a mission project, by Dr. Ning Wang, Director of the Human-Centered AI Lab at USC ICT, Becoming Fei aims to help novice learners build basic competencies in artificial intelligence, particularly machine learning.
More information about the game is available at becomingfei.org. The competition winner will be announced at the I/ITSEC conference in early December.