Timezone: »

Explainable Reinforcement Learning via Model Transforms
Mira Finkelstein · Nitsan levy · Lucy Liu · Yoav Kolumbus · David Parkes · Jeffrey S Rosenschein · Sarah Keren

Thu Dec 01 02:00 PM -- 04:00 PM (PST) @ Hall J #736

Understanding emerging behaviors of reinforcement learning (RL) agents may be difficult since such agents are often trained in complex environments using highly complex decision making procedures. This has given rise to a variety of approaches to explainability in RL that aim to reconcile discrepancies that may arise between the behavior of an agent and the behavior that is anticipated by an observer. Most recent approaches have relied either on domain knowledge, that may not always be available, on an analysis of the agent’s policy, or on an analysis of specific elements of the underlying environment, typically modeled as a Markov Decision Process (MDP). Our key claim is that even if the underlying model is not fully known (e.g., the transition probabilities have not been accurately learned) or is not maintained by the agent (i.e., when using model-free methods), the model can nevertheless be exploited to automatically generate explanations. For this purpose, we suggest using formal MDP abstractions and transforms, previously used in the literature for expediting the search for optimal policies, to automatically produce explanations. Since such transforms are typically based on a symbolic representation of the environment, they can provide meaningful explanations for gaps between the anticipated and actual agent behavior. We formally define the explainability problem, suggest a class of transforms that can be used for explaining emergent behaviors, and suggest methods that enable efficient search for an explanation. We demonstrate the approach on a set of standard benchmarks.

Author Information

Mira Finkelstein (The Hebrew University)
Nitsan levy (Hebrew University of Jerusalem)
Lucy Liu (Harvard University)
Yoav Kolumbus (Hebrew University of Jerusalem)
David Parkes (Harvard University)

David C. Parkes is Gordon McKay Professor of Computer Science in the School of Engineering and Applied Sciences at Harvard University. He was the recipient of the NSF Career Award, the Alfred P. Sloan Fellowship, the Thouron Scholarship and the Harvard University Roslyn Abramson Award for Teaching. Parkes received his Ph.D. degree in Computer and Information Science from the University of Pennsylvania in 2001, and an M.Eng. (First class) in Engineering and Computing Science from Oxford University in 1995. At Harvard, Parkes leads the EconCS group and teaches classes in artificial intelligence, optimization, and topics at the intersection between computer science and economics. Parkes has served as Program Chair of ACM EC’07 and AAMAS’08 and General Chair of ACM EC’10, served on the editorial board of Journal of Artificial Intelligence Research, and currently serves as Editor of Games and Economic Behavior and on the boards of Journal of Autonomous Agents and Multi-agent Systems and INFORMS Journal of Computing. His research interests include computational mechanism design, electronic commerce, stochastic optimization, preference elicitation, market design, bounded rationality, computational social choice, networks and incentives, multi-agent systems, crowd-sourcing and social computing.

Jeffrey S Rosenschein (The Hebrew University of Jerusalem)
Sarah Keren (Technion, Technion)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors