Jul 2026
Top-k Pareto Bandits: Hypervolume Regret for Multi-Objective Slate Selection
An optimistic algorithm is introduced that selects arms greedily based on optimistic estimates of their marginal hypervolume contributions, and a gap-free regret bound and a gap-dependent bound are established that become polylogarithmic in $T$ once the arms are sufficiently well separated.
Nicolas Gutowski, Fabien Chhel, Alexandre Letard et al.
· arXiv.org · 0 citations