Talk from Yu-Xiang Wang “Towards Practical Reinforcement Learning: Offline Data and Low-Adaptive Exploration” (подія в архіві)

“AI for Ukraine” is a series of workshops and lectures held by international artificial intelligence experts to support the development of Ukraine’s tech community during the war. The project was opened by the top-notch speaker from AI and ML global niche Yoshua Bengio (Mila/U. Montreal). The next ones lecturers are Alex J. Smola (Amazon Web), Sebastian Bubeck (Microsoft), Gaël Varoquaux (INRIA), and many other well-known specialists have joined the initiative. This is a non-commercial educational project by AI HOUSE — a company focused on building the AI/ML community in Ukraine and is part of the Roosh tech ecosystem. All proceeds collected upon registration will be donated to the biggest Ukrainian charity fund “Come Back Alive” .

Yu-Xiang Wang, Assistant Professor of Computer Science at UC Santa Barbara, Director of Scalable Statistical Machine Learning Lab.

About the talk: “Towards Practical Reinforcement Learning: Offline Data and Low-Adaptive Exploration”

Standard Reinforcement Learning requires interactive access to the environment for trials-and-errors. In many practical applications, it is often unsafe, illegal or costly to deploy untested policies. Offline RL, instead, directly optimizes the policy from a large logged dataset collected by running the currently deployed systems. This setting is, arguably more common in practical applications.

In this talk, Yu-Xiang Wang will first share some recent theoretical advances on the offline RL algorithms. Then he will talk about the limitation of offline RL over its online counterpart, and describe a new model called RL with low-switching cost that could get the best of both world.

Join the lecture via registration and a free donation here: aiforukraine.aihouse.club

All donations go to the Come Back Alive fund.