Talk by Ostap at AMII University of Alberta

Ostap recently had the opportunity to present at the University of Alberta’s AI Seminar, a renowned series featuring leading AI researchers from around the world, organized by this year's Turing Award Winner, Rich Sutton.

In his talk, Ostap discussed how overestimation bias can critically affect reinforcement learning algorithms, sometimes leading to severe performance drops or even failure. We proposed the T-Estimator (TE), which uses two-sample testing to flexibly manage this bias, and demonstrated how it can be incorporated into Q-Learning and Bootstrapped Deep Q-Networks with provable convergence guarantees.

You can watch the full talk here: https://youtu.be/MhZb5jVPQWU