Adversarially Robust Stochastic Multi-Armed Bandits

Explore the cutting-edge research on adversarially robust stochastic multi-armed bandits in this 45-minute seminar presented by Julian Zimmert at VinAI. Delve into the world of optimal experimental design, with applications in digital advertising and website optimization. Learn about the traditional separation between stochastic and adversarial settings in bandit literature, and discover a recent breakthrough in practical all-purpose algorithms that bridge the gap between fast and robust learning. Gain insights into the challenges faced by practitioners when dealing with real-world applications that fall between purely stochastic and adversarial environments. Understand how hidden confounders and unforeseen changes can impact data distribution and affect the performance of stochastic bandit algorithms. Join this seminar to explore innovative solutions that aim to provide both fast and robust learning in multi-armed bandit problems.