Real-world environments pose many challenges for online decision-making algorithms. In such settings, the feedback gathered through interaction with the environment is often high-dimensional, partially observed, implicit, and often corrupted. Nevertheless, in standard Bandits and Reinforcement Learning (RL) such challenges are often overlooked. This workshop aims to present a broad overview of the contemporary decision-making models that are being actively researched, and provide a networking forum for researchers and practitioners. Our current theme, New Models in Online Decision Making for Real-World Applications, pervades machine learning at large, and with our aim of bridging practice and theory this workshop will find general appeal. The workshop forum will allow us to foster dissemination, cross-fertilization and discussion at scale.

The goal of this workshop is to present recent works that study online decision making in settings that go beyond the standard models. In particular, we are interested in providing a forum for new scenarios in online decision making and algorithms inspired by real-world applications. There is a flurry of yet-to-be studied practically relevant models that fall into this category. The list includes but is not limited to: RL beyond the classical reward-MDP based model, learning in partially observed systems, preference based RL, multiagent games, offline RL, learning with corrupted or delayed feedback, just to name a few. Our overall aim is to showcase interesting works that suggest novel techniques to overcome real world challenges.

The workshop will feature several speakers from computer science, operation research, control, and statistics to highlight recent progress, identify key challenges, and discuss future directions. Contributed talks will be augmented by poster presentations, panel discussions, and virtual social events.

Venue: TTIC Chicago, USA (July 13-15, 2022)

PROGRAM SCHEDULE