CARE: Class-Adaptive Expert Consensus for Reliable Learning with Long-Tailed Noisy Labels

Mengke Li, Haiquan Ling, Lihao Chen, Yang Lu, Yiqun Zhang, Hui Huang
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:68314-68334, 2026.

Abstract

Learning from real-world data is frequently hindered by the compound challenge of long-tailed class distributions and noisy annotations. Existing methods partially address these issues but typically ignore the non-uniform impact of label noise across classes, resulting in ineffective correction for tail classes and over-regularization for head classes. To address this issue, we propose Class-Adaptive Rectification with Experts (CARE), a parameter-efficient framework that leverages three complementary supervision sources from vision-language models (VLM): observed noisy labels, VLM text embeddings, and visual features. CARE introduces a class-adaptive expert consensus mechanism that enforces stricter agreement for tail classes and more permissive agreement for head classes based on class frequency. By aggregating high-confidence predictions across these sources, CARE filters unreliable signals and recalibrates class distributions, yielding more reliable rectification under long-tailed distributions. Extensive experiments on both synthetic and real-world benchmarks demonstrate that CARE consistently outperforms state-of-the-art methods, achieving up to 3.0% performance gains. The source code is available at https://github.com/qwq123-study/CARE.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-li26cy, title = {{CARE}: Class-Adaptive Expert Consensus for Reliable Learning with Long-Tailed Noisy Labels}, author = {Li, Mengke and Ling, Haiquan and Chen, Lihao and Lu, Yang and Zhang, Yiqun and Huang, Hui}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {68314--68334}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/li26cy/li26cy.pdf}, url = {https://proceedings.mlr.press/v306/li26cy.html}, abstract = {Learning from real-world data is frequently hindered by the compound challenge of long-tailed class distributions and noisy annotations. Existing methods partially address these issues but typically ignore the non-uniform impact of label noise across classes, resulting in ineffective correction for tail classes and over-regularization for head classes. To address this issue, we propose Class-Adaptive Rectification with Experts (CARE), a parameter-efficient framework that leverages three complementary supervision sources from vision-language models (VLM): observed noisy labels, VLM text embeddings, and visual features. CARE introduces a class-adaptive expert consensus mechanism that enforces stricter agreement for tail classes and more permissive agreement for head classes based on class frequency. By aggregating high-confidence predictions across these sources, CARE filters unreliable signals and recalibrates class distributions, yielding more reliable rectification under long-tailed distributions. Extensive experiments on both synthetic and real-world benchmarks demonstrate that CARE consistently outperforms state-of-the-art methods, achieving up to 3.0% performance gains. The source code is available at https://github.com/qwq123-study/CARE.} }
Endnote
%0 Conference Paper %T CARE: Class-Adaptive Expert Consensus for Reliable Learning with Long-Tailed Noisy Labels %A Mengke Li %A Haiquan Ling %A Lihao Chen %A Yang Lu %A Yiqun Zhang %A Hui Huang %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-li26cy %I PMLR %P 68314--68334 %U https://proceedings.mlr.press/v306/li26cy.html %V 306 %X Learning from real-world data is frequently hindered by the compound challenge of long-tailed class distributions and noisy annotations. Existing methods partially address these issues but typically ignore the non-uniform impact of label noise across classes, resulting in ineffective correction for tail classes and over-regularization for head classes. To address this issue, we propose Class-Adaptive Rectification with Experts (CARE), a parameter-efficient framework that leverages three complementary supervision sources from vision-language models (VLM): observed noisy labels, VLM text embeddings, and visual features. CARE introduces a class-adaptive expert consensus mechanism that enforces stricter agreement for tail classes and more permissive agreement for head classes based on class frequency. By aggregating high-confidence predictions across these sources, CARE filters unreliable signals and recalibrates class distributions, yielding more reliable rectification under long-tailed distributions. Extensive experiments on both synthetic and real-world benchmarks demonstrate that CARE consistently outperforms state-of-the-art methods, achieving up to 3.0% performance gains. The source code is available at https://github.com/qwq123-study/CARE.
APA
Li, M., Ling, H., Chen, L., Lu, Y., Zhang, Y. & Huang, H.. (2026). CARE: Class-Adaptive Expert Consensus for Reliable Learning with Long-Tailed Noisy Labels. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:68314-68334 Available from https://proceedings.mlr.press/v306/li26cy.html.

Related Material