[日時] 2024年4月12日(金) 13:00--15:00
[場所] 東京大学 本郷キャンパス 工14号館 534室
[講演者] 伊藤 伸志 (東京大学)
[題目(title)] On best-of-both-worlds online learning algorithms
[概要(Abstract)]
In the research areas of online learning and bandit problems, various environment models such as stochastic models and adversarial models have been explored, with algorithms proposed that are suited to each model. Applying these algorithms to real-world problems poses a challenge in selecting the appropriate algorithm. To address this challenge, algorithms have been proposed that automatically adapt to various environment models and work effectively for all models; these are known as best-of-both-worlds algorithms. This talk will introduce recent advances in this stream of research. The topics in this talk include studies presented at COLT22 (https://arxiv.org/abs/2206.06810), NeurIPS22 (https://arxiv.org/abs/2206.00873), and COLT23 (https://arxiv.org/abs/2302.12370).