When: Thursday, July 7, 2022, 12pm (CEST).

Where: Room A2, DIAG & Zoom online, check the address in the Google Calendar Event.

Topic: Factored Planning: From Classical Planning to Dec-POMDPs

Speaker: Prof. Ronen Brafman, Scharf-Ullman Chair in Computer Science at Ben-Gurion University in Israel.

Abstract

The idea of taking a problem and solving it by decomposing it to multiple smaller problems and then somehow combining them together has motivated algorithm design in many different area. I this talk I will review an evolving journey motivated by this idea that I’ve been taking with the help of many colleagues. Starting from an attempt to make classical planning more efficient, it evolves into distributed classical planning, privacy preserving multi-agent planning, multi-agent forward search, and eventually (for now?) planning in decentralized POMDPs. Common to most of these techniques is the idea of identifying the parts of the problem (variables/actions) involving multiple components/agents and those relevant to one component only. This then allows us to formulate an abstract problem whose solution forms a skeleton to the solution of the real problem. This skeleton then serves to partition the problem into multiple independent problems involving one component/agent only, that can be solved in parallel.

Short Bio

Prof. Ronen Brafman is the Scharf-Ullman Chair in Computer Science at Ben-Gurion University in Israel. He obtained his Ph.D. from Stanford University in 1996, was a Post-Doctoral Fellow at UBC, and was a research scientist at NASA AMES Research Center between 2004-2006. His work spans diverse aspects of decision making, including preference modeling and reinforcement learning, where he co-developed the CP-nets model and the R-max algorithm, but he is mostly interested in decision making under uncertainty, and more recently, planning for multi-agent systems where he co-introduced the MA-STRIPS model and privacy preserving planning. Ronen is also interested in applying planning to robotics, and improving the practice of software engineering for autonomous robots. He served as Program and Conference Chair for ICAPS, an Associate Editor for JAIR, AIJ, and JAAMAS, and is a AAAI and EurAI fellow.

Material