Learning Infinitely Many Coins Simultaneously

This Hebrew lecture by Professor Aryeh Kontorovich explores the complex problem of learning infinitely many coin biases simultaneously. Delve into the mathematical challenges of estimating multiple Bernoulli parameters beyond the well-understood cases of single coins or finite sets of coins. Discover why estimating infinite sequences of parameters is impossible in the worst case, and learn about the conditions under which the expected maximum deviation between true and estimated values can converge to zero. The talk presents exhaustive answers for independent or negatively dependent Bernoullis, while highlighting the additional complexities introduced by positive dependencies. Professor Kontorovich, a full professor at Ben-Gurion University with expertise in machine learning, probability, and statistics, will present this joint work with Moïse Blanchard, Doron Cohen, and Václav Voráček. The lecture takes place on Thursday, March 20th, 2025, at 10:30 AM in room B220, organized by the HUJI Machine Learning Club.