Reproduce “Balanced Sampling in Stochastic Multi-objective 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
Mengfan Xu
The authors propose methods to balance sampling across unknown Pareto-optimal arms while learning which belong to that set. Their terminal unfairness measures divergence from an even allocation, penalizing dominated pulls and imbalance among Pareto arms. Under positive-separation and sufficient-horizon conditions, ATT achieves expected unfairness O(log T/T); EG-ATT retains that rate and has variance O(log² T/T²). These guarantees exclude zero-separation instances.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.