Jorge K. S. Kamassury;Henrique Pickler;Filipe R. Cordeiro;Danilo Silva
{"title":"CCT: A Cyclic Co-Teaching Approach to Train Deep Neural Networks With Noisy Labels","authors":"Jorge K. S. Kamassury;Henrique Pickler;Filipe R. Cordeiro;Danilo Silva","doi":"10.1109/ACCESS.2025.3548510","DOIUrl":null,"url":null,"abstract":"The detrimental impact of noisy labels on the generalization performance of deep neural networks has sparked research interest in learning with noisy labels (LNL). Among the various methods proposed to mitigate this effect, the Co-Teaching method, characterized by co-training with the small-loss criterion, is one of the most established approaches and is widely employed as a key component in recent LNL methods. Although Co-Teaching can mitigate the overfitting effect, it still remains, especially in scenarios with high rates of label noise in datasets. Strategies from the LNL literature to address this typically include the use of disagreement techniques and alternative loss functions. In this paper, we propose the Cyclic Co-Teaching (CCT) method, which employs cyclic variations in the learning rate and sample retention rate at the mini-batch level, along with a checkpoint mechanism that ensures that training in subsequent cycles always resumes from the best models obtained so far. For optimizing the method, we developed a framework that incorporates a pre-training phase to obtain an optimized vanilla model used to initialize CCT model weights, and a transparent univariate optimization strategy for hyperparameters that does not necessarily require a clean validation set. Experimental results on synthetic and real-world datasets, under different types and levels of noise and employing various neural network architectures, demonstrate that CCT outperforms several state-of-the-art LNL methods in most evaluated scenarios.","PeriodicalId":13079,"journal":{"name":"IEEE Access","volume":"13 ","pages":"43843-43860"},"PeriodicalIF":3.4000,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10912480","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE Access","FirstCategoryId":"94","ListUrlMain":"https://ieeexplore.ieee.org/document/10912480/","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
引用次数: 0
Abstract
The detrimental impact of noisy labels on the generalization performance of deep neural networks has sparked research interest in learning with noisy labels (LNL). Among the various methods proposed to mitigate this effect, the Co-Teaching method, characterized by co-training with the small-loss criterion, is one of the most established approaches and is widely employed as a key component in recent LNL methods. Although Co-Teaching can mitigate the overfitting effect, it still remains, especially in scenarios with high rates of label noise in datasets. Strategies from the LNL literature to address this typically include the use of disagreement techniques and alternative loss functions. In this paper, we propose the Cyclic Co-Teaching (CCT) method, which employs cyclic variations in the learning rate and sample retention rate at the mini-batch level, along with a checkpoint mechanism that ensures that training in subsequent cycles always resumes from the best models obtained so far. For optimizing the method, we developed a framework that incorporates a pre-training phase to obtain an optimized vanilla model used to initialize CCT model weights, and a transparent univariate optimization strategy for hyperparameters that does not necessarily require a clean validation set. Experimental results on synthetic and real-world datasets, under different types and levels of noise and employing various neural network architectures, demonstrate that CCT outperforms several state-of-the-art LNL methods in most evaluated scenarios.
IEEE AccessCOMPUTER SCIENCE, INFORMATION SYSTEMSENGIN-ENGINEERING, ELECTRICAL & ELECTRONIC
CiteScore
9.80
自引率
7.70%
发文量
6673
审稿时长
6 weeks
期刊介绍:
IEEE Access® is a multidisciplinary, open access (OA), applications-oriented, all-electronic archival journal that continuously presents the results of original research or development across all of IEEE''s fields of interest.
IEEE Access will publish articles that are of high interest to readers, original, technically correct, and clearly presented. Supported by author publication charges (APC), its hallmarks are a rapid peer review and publication process with open access to all readers. Unlike IEEE''s traditional Transactions or Journals, reviews are "binary", in that reviewers will either Accept or Reject an article in the form it is submitted in order to achieve rapid turnaround. Especially encouraged are submissions on:
Multidisciplinary topics, or applications-oriented articles and negative results that do not fit within the scope of IEEE''s traditional journals.
Practical articles discussing new experiments or measurement techniques, interesting solutions to engineering.
Development of new or improved fabrication or manufacturing techniques.
Reviews or survey articles of new or evolving fields oriented to assist others in understanding the new area.