Coordinated Sparse Recovery of Label Noise

arXiv:2404.04800v1 Announce Type: new
Abstract: Label noise is a common issue in real-world datasets that inevitably impacts the generalization of models. This study focuses on robust classification tasks where the label noise is instance-dependent. Estimating the transition matrix accurately in this task is challenging, and methods based on sample selection often exhibit confirmation bias to varying degrees. Sparse over-parameterized training (SOP) has been theoretically effective in estimating and recovering label noise, offering a novel solution for noise-label learning. However, this …

abstract arxiv bias classification confirmation bias cs.cv cs.lg datasets impacts instance issue matrix noise recovery robust sample stat.ml study tasks transition type world