English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Label noise in training data can significantly degrade a model'sgeneralization performance for supervised learning tasks. Here we focus on theproblem that noisy labels are primarily mislabeled samples, which tend to beconcentrated near decision boundaries, rather than uniformly distributed, andwhose features should be equivocal. To address the problem, we propose anensemble learning method to correct noisy labels by exploiting the localstructures of feature manifolds. Different from typical ensemble strategiesthat increase the prediction diversity among sub-models via certain loss terms,our method trains sub-models on disjoint subsets, each being a union of thenearest-neighbors of randomly selected seed samples on the data manifold. As aresult, each sub-model can learn a coarse representation of the data manifoldalong with a corresponding graph. Moreover, only a limited number of sub-modelswill be affected by locally-concentrated noisy labels. The constructed graphsare used to suggest a series of label correction candidates, and accordingly,our method derives label correction results by voting down inconsistentsuggestions. Our experiments on real-world noisy label datasets demonstrate thesuperiority of the proposed method over existing state-of-the-arts.

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