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

Noisy labels damage the performance of deep networks. For robust learning, aprominent two-stage pipeline alternates between eliminating possible incorrectlabels and semi-supervised training. However, discarding part of noisy labelscould result in a loss of information, especially when the corruption has adependency on data, e.g., class-dependent or instance-dependent. Moreover, fromthe training dynamics of a representative two-stage method DivideMix, weidentify the domination of confirmation bias: pseudo-labels fail to correct aconsiderable amount of noisy labels, and consequently, the errors accumulate.To sufficiently exploit information from noisy labels and mitigate wrongcorrections, we propose Robust Label Refurbishment (Robust LR) a new hybridmethod that integrates pseudo-labeling and confidence estimation techniques torefurbish noisy labels. We show that our method successfully alleviates thedamage of both label noise and confirmation bias. As a result, it achievesstate-of-the-art performance across datasets and noise types, namely CIFARunder different levels of synthetic noise and Mini-WebVision and ANIMAL-10Nwith real-world noise.

Two Wrongs Don't Make a Right: Combating Confirmation Bias in Learning with Label Noise