Global Anchor Consensus Boosts Multi-View Classification with Noisy Labels.

Researchers have developed a new method called Global Anchor-based Label Auditing (GALA) to improve robust multi-view classification, particularly when dealing with noisy labels. Existing multi-view methods often struggle with imperfect annotations, as their refinement signals can become unreliable. GALA addresses this by creating a stable "global anchor" for each class within every view. These global anchors aggregate samples of an entire class, providing a robust reference point that is less sensitive to individual noisy predictions. Each view then assesses how close an instance is to its observed label's anchor relative to competing anchors. These per-view evaluations are combined with classifier confidence to form a cross-view audit score. Based on these audit scores, suspicious samples are assigned lower weights, and an adaptive correction strategy rewrites labels only when the anchor-based candidate aligns with the classifier's prediction. This iterative process refines both the anchors and the noise-robust representation learning. Extensive experiments on six datasets demonstrate GALA's superior performance, especially under high noise rates, compared to eight state-of-the-art methods.