Reproduce “A conversion theorem and minimax optimality for continuum contextual bandits” 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
Arya Akhavan, Karim Lounici, Massimiliano Pontil, Alexandre B. Tsybakov
The authors convert a static-regret bandit algorithm into local learners for contextual continuous-action decisions, grouping nearby contexts and balancing approximation error against the cost of multiple learners. Under Hölder-smooth losses, the method gives sublinear contextual regret, with rates minimax-optimal in the query horizon up to logarithmic factors in several settings. Without continuity, linear regret may be unavoidable.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.