Variational Hidden Conditional Random Fields with Coupled Dirichlet Process Mixtures (bibtex)

by Bousmalis, Konstantinos, Zafeiriou, Stefanos, Morency, Louis–Philippe, Pantic, Maja and Ghahramani, Zoubin

Abstract:

Hidden Conditional Random Fields (HCRFs) are discriminative latent variable models which have been shown to successfully learn the hidden structure of a given classiﬁcation problem. An inﬁnite HCRF is an HCRF with a countably inﬁnite number of hidden states, which rids us not only of the necessity to specify a priori a ﬁxed number of hidden states available but also of the problem of overﬁtting. Markov chain Monte Carlo (MCMC) sampling algorithms are often employed for inference in such models. However, convergence of such algorithm is rather difficult to verify, and as the complexity of the task at han increases, the computational cost of such algorithms often becomes prohibitive. These limitations can be overcome by variational techniques. In this paper, we present a generalized framework for inﬁnite HCRF models, and a novel variational inference approach on a model based on coupled Dirichlet Process Mixtures, the HCRF–DPM. We show that the variational HCRF–DPM is able to converge to a correct number of represented hidden states, and performs as well as the best parametric HCRFs—chosen via cross-validation— for the difficult tasks of recognizing instances of agreement, disagreement, and pain in audiovisual sequences.

Reference:

Variational Hidden Conditional Random Fields with Coupled Dirichlet Process Mixtures (Bousmalis, Konstantinos, Zafeiriou, Stefanos, Morency, Louis–Philippe, Pantic, Maja and Ghahramani, Zoubin), In Machine Learning and Knowledge Discovery in Databases (Hutchison, David, Kanade, Takeo, Kittler, Josef, Kleinberg, Jon M., Mattern, Friedemann, Mitchell, John C., Naor, Moni, Nierstrasz, Oscar, Pandu Rangan, C., Steffen, Bernhard, Sudan, Madhu, Terzopoulos, Demetri, Tygar, Doug, Vardi, Moshe Y., Weikum, Gerhard, Blockeel, Hendrik, Kersting, Kristian, Nijssen, Siegfried, Železný, Filip, eds.), Springer Berlin Heidelberg, volume 8189, 2013.

Bibtex Entry:

@inproceedings{hutchison_variational_2013, address = {Prague, Czech Republic}, title = {Variational {Hidden} {Conditional} {Random} {Fields} with {Coupled} {Dirichlet} {Process} {Mixtures}}, volume = {8189}, isbn = {978-3-642-40990-5 978-3-642-40991-2}, url = {http://link.springer.com/10.1007/978-3-642-40991-2_34}, doi = {10.1007/978-3-642-40991-2_34}, abstract = {Hidden Conditional Random Fields (HCRFs) are discriminative latent variable models which have been shown to successfully learn the hidden structure of a given classiﬁcation problem. An inﬁnite HCRF is an HCRF with a countably inﬁnite number of hidden states, which rids us not only of the necessity to specify a priori a ﬁxed number of hidden states available but also of the problem of overﬁtting. Markov chain Monte Carlo (MCMC) sampling algorithms are often employed for inference in such models. However, convergence of such algorithm is rather difficult to verify, and as the complexity of the task at han increases, the computational cost of such algorithms often becomes prohibitive. These limitations can be overcome by variational techniques. In this paper, we present a generalized framework for inﬁnite HCRF models, and a novel variational inference approach on a model based on coupled Dirichlet Process Mixtures, the HCRF–DPM. We show that the variational HCRF–DPM is able to converge to a correct number of represented hidden states, and performs as well as the best parametric HCRFs—chosen via cross-validation— for the difficult tasks of recognizing instances of agreement, disagreement, and pain in audiovisual sequences.}, booktitle = {Machine {Learning} and {Knowledge} {Discovery} in {Databases}}, publisher = {Springer Berlin Heidelberg}, author = {Bousmalis, Konstantinos and Zafeiriou, Stefanos and Morency, Louis–Philippe and Pantic, Maja and Ghahramani, Zoubin}, editor = {Hutchison, David and Kanade, Takeo and Kittler, Josef and Kleinberg, Jon M. and Mattern, Friedemann and Mitchell, John C. and Naor, Moni and Nierstrasz, Oscar and Pandu Rangan, C. and Steffen, Bernhard and Sudan, Madhu and Terzopoulos, Demetri and Tygar, Doug and Vardi, Moshe Y. and Weikum, Gerhard and Blockeel, Hendrik and Kersting, Kristian and Nijssen, Siegfried and Železný, Filip}, month = sep, year = {2013}, keywords = {Virtual Humans, UARC}, pages = {531--547} }

Powered by bibtexbrowser