Reproduce “DOPL: Direct Online Preference Learning for Restless Bandits with Preference Feedback” 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
Guojun Xiong, Ujwal Dinesha, Debajoy Mukherjee, Jian Li, Srinivas Shakkottai
DOPL learns allocation in restless multi-armed bandits from pairwise preferences rather than numeric rewards. It assumes Bradley–Terry comparisons, estimates preferences online, and uses an optimistic index policy with transition confidence sets. In simulation, its reward trace approaches the full-information oracle and exceeds plotted learning baselines. Under stated assumptions, it guarantees regret of order tilde-O of the square root of T log T relative to the offline index policy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.