AIC Seminar Series

In this talk, I will discuss two projects in the MAPLE Laboratory at UMBC that are focusing on modeling and learning user preferences for a variety of applications.

The first part of the talk will describe the methods that we are developing for learning planning preferences in the context of DARPA’s Integrated Learning program. Given a single planning trace, we show how to learn a preference function that models the user’s preferences for alternative variable bindings (e.g., which destination hospital to select when transporting a wounded soldier). The learned preference function is represented as a lexicographic ordering that characterizes the relative importance of the object attributes. I will also mention previous work that we have done in this domain on extending CP-nets (conditional preference networks) to handle preferences over continuous attributes.

The second part of the talk will describe an NSF-funded effort in which we have developed a preference language and learning algorithms for modeling preferences over sets of objects. These preferences are useful in contexts where the system is asked to retrieve a collection of relevant objects from a larger pool. We are currently applying these methods to the problem of selecting images to download from a Mars rover. I will conclude by discussing other possible applications, and how these two projects could be merged into a more general framework for learning preferences for decision making.

Joint work with Fusun Yaman (UMBC) and Kiri Wagstaff (JPL)

Dr. Marie desJardins is an Associate Professor of Computer Science at the University of Maryland, Baltimore County. She received her Ph.D. in 1992 from the University of California, Berkeley. Her research is in artificial intelligence, primarily in the areas of multi-agent systems, planning, and multi-agent systems, with a focus on interactive AI and applications of AI. Prior to joining the faculty at UMBC, she was a Senior Computer Scientist in the AI Center at SRI International.