Meeting Description: Agents exploring and interacting with an unknown environment is a common challenge in many scientific disciplines. It is particularly challenging when actions taken early on have a strong affect on the set of future possibilities. In these scenarios, taking actions that only maximize the immediate reward can lead to poor outcomes. Instead, agents must use foresight to understand the full consequences of their current actions, that is, they must look ahead and predict how they will act in the future based on how they act today. This workshop seeks to bring together diverse scientific groups in an attempt to unify thought on lookahead policies in optimization. In particular, it seeks to connect mathematical results for particular mathematical structures, with behavioral results showing agents effectively navigating complex environments, with engineers faced with difficult planning problems.
Collins Conference Room
This event is by invitation only.
Mirta Galesic, Brendan Tracey, and David Wolpert