Export Reviews, Discussions, Author Feedback and Meta-Reviews

This paper presents an EM method for solving interactive POMDPs (I-POMDPS), which exploits problem structure in the I-POMDP model. Specifically, an EM method for I-POMDPs is introduced, along with improvements which use block-coordinate descent and forward filtering-backward sampling. Experimental results show significant scalability gains using some of these methods. To the best of my knowledge, this is the first EM method applied to I-POMDPs. While I-POMDPs have many similarities to POMDP (and Dec-POMDPs), where EM has been used, there is additional structure in I-POMDPs in the form of models of the other agents in the problem.