arxivcs.LGstat.APstat.MLstat.OT2026-07-09
Optimal Top-$k$ Identification from Pairwise Comparisons
We study the active learning problem of fixed-confidence top-$k$ identification from noisy pairwise comparisons. In this problem, an algorithm sequentially chooses pairs of items to compare, observes the outcomes, and stops when it can return the set of top-$k$ items with error p…