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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Label noise in real-world datasets encodes wrong correlation patterns andimpairs the generalization of deep neural networks (DNNs). It is critical tofind efficient ways to detect corrupted patterns. Current methods primarilyfocus on designing robust training techniques to prevent DNNs from memorizingcorrupted patterns. These approaches often require customized trainingprocesses and may overfit corrupted patterns, leading to a performance drop indetection. In this paper, from a more data-centric perspective, we propose atraining-free solution to detect corrupted labels. Intuitively, ``closer''instances are more likely to share the same clean label. Based on theneighborhood information, we propose two methods: the first one uses ``localvoting" via checking the noisy label consensuses of nearby features. The secondone is a ranking-based approach that scores each instance and filters out aguaranteed number of instances that are likely to be corrupted. Wetheoretically analyze how the quality of features affects the local voting andprovide guidelines for tuning neighborhood size. We also prove the worst-caseerror bound for the ranking-based method. Experiments with both synthetic andreal-world label noise demonstrate our training-free solutions consistently andsignificantly improve most of the training-based baselines. Code is availableat github.com/UCSC-REAL/SimiFeat.

Detecting Corrupted Labels Without Training a Model to Predict