Detecting Corrupted Labels Without Training a Model to Predict

Zhaowei Zhu, Zihao Dong, Yang Liu
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:27412-27427, 2022.

Abstract

Label noise in real-world datasets encodes wrong correlation patterns and impairs the generalization of deep neural networks (DNNs). It is critical to find efficient ways to detect corrupted patterns. Current methods primarily focus on designing robust training techniques to prevent DNNs from memorizing corrupted patterns. These approaches often require customized training processes and may overfit corrupted patterns, leading to a performance drop in detection. In this paper, from a more data-centric perspective, we propose a training-free solution to detect corrupted labels. Intuitively, “closer” instances are more likely to share the same clean label. Based on the neighborhood information, we propose two methods: the first one uses “local voting" via checking the noisy label consensuses of nearby features. The second one is a ranking-based approach that scores each instance and filters out a guaranteed number of instances that are likely to be corrupted. We theoretically analyze how the quality of features affects the local voting and provide guidelines for tuning neighborhood size. We also prove the worst-case error bound for the ranking-based method. Experiments with both synthetic and real-world label noise demonstrate our training-free solutions consistently and significantly improve most of the training-based baselines. Code is available at github.com/UCSC-REAL/SimiFeat.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-zhu22a, title = {Detecting Corrupted Labels Without Training a Model to Predict}, author = {Zhu, Zhaowei and Dong, Zihao and Liu, Yang}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {27412--27427}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/zhu22a/zhu22a.pdf}, url = {https://proceedings.mlr.press/v162/zhu22a.html}, abstract = {Label noise in real-world datasets encodes wrong correlation patterns and impairs the generalization of deep neural networks (DNNs). It is critical to find efficient ways to detect corrupted patterns. Current methods primarily focus on designing robust training techniques to prevent DNNs from memorizing corrupted patterns. These approaches often require customized training processes and may overfit corrupted patterns, leading to a performance drop in detection. In this paper, from a more data-centric perspective, we propose a training-free solution to detect corrupted labels. Intuitively, “closer” instances are more likely to share the same clean label. Based on the neighborhood information, we propose two methods: the first one uses “local voting" via checking the noisy label consensuses of nearby features. The second one is a ranking-based approach that scores each instance and filters out a guaranteed number of instances that are likely to be corrupted. We theoretically analyze how the quality of features affects the local voting and provide guidelines for tuning neighborhood size. We also prove the worst-case error bound for the ranking-based method. Experiments with both synthetic and real-world label noise demonstrate our training-free solutions consistently and significantly improve most of the training-based baselines. Code is available at github.com/UCSC-REAL/SimiFeat.} }
Endnote
%0 Conference Paper %T Detecting Corrupted Labels Without Training a Model to Predict %A Zhaowei Zhu %A Zihao Dong %A Yang Liu %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-zhu22a %I PMLR %P 27412--27427 %U https://proceedings.mlr.press/v162/zhu22a.html %V 162 %X Label noise in real-world datasets encodes wrong correlation patterns and impairs the generalization of deep neural networks (DNNs). It is critical to find efficient ways to detect corrupted patterns. Current methods primarily focus on designing robust training techniques to prevent DNNs from memorizing corrupted patterns. These approaches often require customized training processes and may overfit corrupted patterns, leading to a performance drop in detection. In this paper, from a more data-centric perspective, we propose a training-free solution to detect corrupted labels. Intuitively, “closer” instances are more likely to share the same clean label. Based on the neighborhood information, we propose two methods: the first one uses “local voting" via checking the noisy label consensuses of nearby features. The second one is a ranking-based approach that scores each instance and filters out a guaranteed number of instances that are likely to be corrupted. We theoretically analyze how the quality of features affects the local voting and provide guidelines for tuning neighborhood size. We also prove the worst-case error bound for the ranking-based method. Experiments with both synthetic and real-world label noise demonstrate our training-free solutions consistently and significantly improve most of the training-based baselines. Code is available at github.com/UCSC-REAL/SimiFeat.
APA
Zhu, Z., Dong, Z. & Liu, Y.. (2022). Detecting Corrupted Labels Without Training a Model to Predict. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:27412-27427 Available from https://proceedings.mlr.press/v162/zhu22a.html.

Related Material