Reproduce “Best-of-Both-Worlds FTRL for Linear Bandits: The Interplay of Information Acquisition, Corruption, and Geometry” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Canzhe Zhao, Vincent Y. F. Tan
IA-FTRL is a purely FTRL-based method for finite-action linear bandits that adapts exploration to estimated action gaps rather than switching regimes. It guarantees adversarial regret O(d log K + √(dT log K)) and corrupted-stochastic regret O(c* log K log T + M + J√C), where M is independent of T and C. The corruption coefficient still depends on the minimum gap, and the stochastic bound includes an extra log K factor.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.