Cover First, Disagree Softly: Rethinking Mismatch-First Active Learning for Frame-Level Audio Classification

arXiv cs.AI · 18 hours ago

Researchers propose mismatch-weighted facility location (MW-FL), a new active learning strategy for frame-level audio classification that replaces the mismatch-first farthest-traversal approach with a disagreement-weighted coverage objective. The method showed improvements across two datasets by penalizing similarity among selected segments rather than using hard sequential decisions to prioritize high-disagreement segments first. The new approach achieved the best area under the learning curve on both evaluated datasets by treating coverage as the dominant factor in segment selection.