Missing Mass and Optimal Discovery in Power Systems Security Analysis

Watch a 57-minute lecture by Prof. Aurélien Garivier from Ecole Normale Supérieure de Lyon exploring a novel security analysis problem in power systems called "optimal discovery with probabilistic expert advice." Learn about an innovative algorithm that combines the optimistic paradigm with the Good-Turing missing mass estimator, examining its performance through two distinct regret bounds under weak probabilistic expert assumptions. Discover how implementing stricter assumptions demonstrates macroscopic optimality when compared to oracle strategy and uniform sampling approaches. Examine numerical experiments that validate the theoretical framework and illustrate the algorithm's real-world performance. Benefit from Prof. Garivier's extensive expertise in stochastic and statistical modeling, particularly his influential work in bandit models, Markov models, perfect simulation, and stochastic function optimization, as well as his recent research in differential privacy and risk-aware reinforcement learning.