Reproduce “Approximate information maximization for bandit games” 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
Alex Barbier-Chebbah, Christian L. Vestergaard, Jean-Baptiste Masson, Etienne Boursier
AIM addresses multi-armed bandits by estimating uncertainty about the value of the best arm’s mean reward. It approximates the entropy of that value and selects the arm expected to reduce it. In a fixed-mean Gaussian two-arm simulation with means 0.80 and 0.79, AIM has lower regret than Thompson sampling at longer times. Its asymptotic regret guarantee applies to Gaussian rewards with common variance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.