Ensemble Learning With Manifold-Based Data Splitting for Noisy Label Correction

Label noise in training data can significantly degrade a model’s generalization performance for supervised learning tasks. Here we focus on the problem that noisy labels are primarily caused by mislabeled confusing samples, which tend to be concentrated near decision boundaries rather than uniformly distributed, and whose features should equivocal. To address problem, propose an ensemble method correct exploiting local structures of feature manifolds. Different from typical strategies increase prediction diversity among sub-models via certain loss terms, our trains disjoint subsets, each being union randomly selected seed samples’ nearest-neighbors same class manifold. As result, only limited number will affected locally-concentrated labels, sub-model learn coarse representation manifold along with corresponding graph. The constructed graphs used suggest set label correction candidates, accordingly, determines results majority decisions. Our experiments real-world datasets demonstrate superiority proposed over existing state-of-the-arts.