AIC Seminar Series

There have been several personal assistant systems ranging from desktop assistants to the ones that assist the elderly. These are well engineered solutions to specific domains and lack an overarching framework. In the first half of the talk, I introduce a decision-theoretic model that captures the general notion of assistance where the goal of the assistant is to minimize the cost of user actions. I present the results in a folder predictor domain where we compare our results against a cost-sensitive supervised learning algorithm.

In the second half of the talk, I extend this work to domains where the user’s policies have rich relational and hierarchical structure. Our results indicate that relational hierarchies allow succinct encoding of prior knowledge for the assistant, which in turn enables the assistant to start helping the user after a relatively small amount of experience.

Sriraam Natarajan is a PhD candidate working with Prof.Prasad Tadepalli at Oregon State University. He recieved his Bachelor’s from the University of Madras in 2001 and his M.S from Oregon State University in 2004. His PhD thesis work focusses on developing relational models for intelligent assistants. He has published papers on learning in statistical relational models, reinforcement learning and combining the relational models with decision-theory. He is currently a summer intern at SRI focussing on developing a relational model for activity recognition in CALO.