Stephanie Rosenthal, Dan Bohus, Ece Kamar, and Eric Horvitz
A key decision facing autonomous systems with ac- cess to streams of sensory data is whether to act based on current evidence or to wait for additional information that might enhance the utility of tak- ing an action. Computing the value of informa- tion is particularly difﬁcult with streaming high- dimensional sensory evidence. We describe a belief projection approach to reasoning about information value in these settings, using models for inferring future beliefs over states given streaming evidence. These belief projection models can be learned from data or constructed via direct assessment of param- eters and they ﬁt naturally in modular, hierarchical state inference architectures. We describe princi- ples of using belief projection and present results drawn from an implementation of the methodology within a conversational system.
|Publisher||International Joint Conference on Artificial Intelligence|