"Bias–Tail Concentration Tradeoffs for Stochastic Approximation Algorithms"

Martin Zubeldia

• Assistant Professor, Department of Industrial and Systems Engineering.

Session Information:

📅 Time: April 28th (Tuesday) |  5:00PM - 6:00PM
📍 Location: Lind Hall 204 (In person)  

About the session:

In this talk we will introduce the modeling of Reinforcement Learning algorithms as stochastic iterations that approximately solve a fixed point equation. Then, we will show under what circumstances the stochastic error in these algorithms is necessarily heavy tailed, and what can be done to lighten these tails.