ANNE: Adaptive Nearest Neighbours and Eigenvector-based sample selection for robust learning with noisy labels

Filipe R. Cordeiro, Gustavo Carneiro

Abstract: Highlights•New sample selection for noisy-label learning by combining benefits of AKNN and FINE.•AKNN adapts k-value to noise rates, selecting low-confidence clean-label samples.•ANNE sample selection integrates with SOTA SSR+, validated on real-world datasets.