https://aktumu.github.io/tag/online-learning/ Online Learning Wowchemy (https://wowchemy.com)en-usFri, 01 Dec 2023 00:00:00 +0000 https://aktumu.github.io/media/icon_huf81513cd465b0045e76ded67516f9fbb_13124_512x512_fill_lanczos_center_3.png https://aktumu.github.io/tag/online-learning/ Fri, 01 Dec 2023 00:00:00 +0000 <p>We study online linear optimization as a principled framework for best-of-both-worlds (BOBW) online learning, i.e., attaining the optimal regret rate for the stochastic regime whenever data is iid and the optimal regret rate for the adversarial regime whenever data is adversarial without requiring any prior knowledge about the data generating process. In particular, we demonstrate the surprising versatility of online mirror descent with time-varying Legendre regularizers in the context of BOBW prediction with expert advice (full information setting) and multi-armed bandits (partial information setting).</p>