Cyclic Learning Rate-Based Co-Training for Image Classification With Noisy Labels
Deep learning has excelled in image classification, but noisy labels in large datasets pose a significant challenge, impacting performance and generalization. To tackle this, we propose a novel co-training method using cyclic learning rates. This method trains two networks simultaneously, each selec...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10829578/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1841542567724843008 |
|---|---|
| author | Ying Zheng Yu Gu Pingping Bai Dong Yuan Siqi Zhou Xin Lyu Ang Chen |
| author_facet | Ying Zheng Yu Gu Pingping Bai Dong Yuan Siqi Zhou Xin Lyu Ang Chen |
| author_sort | Ying Zheng |
| collection | DOAJ |
| description | Deep learning has excelled in image classification, but noisy labels in large datasets pose a significant challenge, impacting performance and generalization. To tackle this, we propose a novel co-training method using cyclic learning rates. This method trains two networks simultaneously, each selecting clean samples based on loss values to optimize the other’s parameters, reducing overfitting and confirmation bias. The cyclic learning rate allows networks to oscillate between underfitting and overfitting, enhancing the distinction between clean and noisy samples. Our approach improves noise detection accuracy and robustness against label noise on datasets like CIFAR-10, CIFAR-100, and Clothing1M. Especially on CIFAR-10 and CIFAR-100 with 40% symmetric noise ratio, and Clothing1M, it outperforms the most relevant O2U-Net by 2.59%, 6.11%, and 0.57% in test accuracy, respectively, demonstrating superior noise resistance and classification accuracy under various noise conditions. Comprehensive experiments confirm the effectiveness of our method, advancing image classification in the presence of noisy labels. |
| format | Article |
| id | doaj-art-939dc708791f48d8b28af6c13b47012c |
| institution | Kabale University |
| issn | 2169-3536 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj-art-939dc708791f48d8b28af6c13b47012c2025-01-14T00:02:38ZengIEEEIEEE Access2169-35362025-01-01136292630510.1109/ACCESS.2025.352633210829578Cyclic Learning Rate-Based Co-Training for Image Classification With Noisy LabelsYing Zheng0https://orcid.org/0009-0007-9993-7643Yu Gu1https://orcid.org/0009-0005-1941-6668Pingping Bai2https://orcid.org/0009-0001-7665-4911Dong Yuan3https://orcid.org/0009-0005-8167-369XSiqi Zhou4Xin Lyu5https://orcid.org/0000-0003-1862-2070Ang Chen6https://orcid.org/0009-0003-0406-5936State Key Laboratory of Hydraulic Engineering Intelligent Construction and Operation, Tianjin University, Tianjin, ChinaShandong Province Water Transfer Project Operation and Maintenance Center, Jinan, ChinaShandong Province Water Transfer Project Operation and Maintenance Center, Jinan, ChinaShandong Water Conservancy Survey and Design Institute Company Ltd., Jinan, ChinaCollege of Computer and Information, Hohai University, Nanjing, ChinaCollege of Computer and Information, Hohai University, Nanjing, ChinaCollege of Computer and Information, Hohai University, Nanjing, ChinaDeep learning has excelled in image classification, but noisy labels in large datasets pose a significant challenge, impacting performance and generalization. To tackle this, we propose a novel co-training method using cyclic learning rates. This method trains two networks simultaneously, each selecting clean samples based on loss values to optimize the other’s parameters, reducing overfitting and confirmation bias. The cyclic learning rate allows networks to oscillate between underfitting and overfitting, enhancing the distinction between clean and noisy samples. Our approach improves noise detection accuracy and robustness against label noise on datasets like CIFAR-10, CIFAR-100, and Clothing1M. Especially on CIFAR-10 and CIFAR-100 with 40% symmetric noise ratio, and Clothing1M, it outperforms the most relevant O2U-Net by 2.59%, 6.11%, and 0.57% in test accuracy, respectively, demonstrating superior noise resistance and classification accuracy under various noise conditions. Comprehensive experiments confirm the effectiveness of our method, advancing image classification in the presence of noisy labels.https://ieeexplore.ieee.org/document/10829578/Deep learningcyclic learning rateco-trainingimage classificationnoisy label |
| spellingShingle | Ying Zheng Yu Gu Pingping Bai Dong Yuan Siqi Zhou Xin Lyu Ang Chen Cyclic Learning Rate-Based Co-Training for Image Classification With Noisy Labels IEEE Access Deep learning cyclic learning rate co-training image classification noisy label |
| title | Cyclic Learning Rate-Based Co-Training for Image Classification With Noisy Labels |
| title_full | Cyclic Learning Rate-Based Co-Training for Image Classification With Noisy Labels |
| title_fullStr | Cyclic Learning Rate-Based Co-Training for Image Classification With Noisy Labels |
| title_full_unstemmed | Cyclic Learning Rate-Based Co-Training for Image Classification With Noisy Labels |
| title_short | Cyclic Learning Rate-Based Co-Training for Image Classification With Noisy Labels |
| title_sort | cyclic learning rate based co training for image classification with noisy labels |
| topic | Deep learning cyclic learning rate co-training image classification noisy label |
| url | https://ieeexplore.ieee.org/document/10829578/ |
| work_keys_str_mv | AT yingzheng cycliclearningratebasedcotrainingforimageclassificationwithnoisylabels AT yugu cycliclearningratebasedcotrainingforimageclassificationwithnoisylabels AT pingpingbai cycliclearningratebasedcotrainingforimageclassificationwithnoisylabels AT dongyuan cycliclearningratebasedcotrainingforimageclassificationwithnoisylabels AT siqizhou cycliclearningratebasedcotrainingforimageclassificationwithnoisylabels AT xinlyu cycliclearningratebasedcotrainingforimageclassificationwithnoisylabels AT angchen cycliclearningratebasedcotrainingforimageclassificationwithnoisylabels |