Research map / Label-efficient and robust learning / Noisy labels
Class-Independent Regularization for Learning with Noisy Labels
Rumeng Yi, Dayan Guan, Yaping Huang, Shijian Lu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use class-independent multi-binary classifiers for sample selection, coordinated with a softmax classifier for classification | CIR · CIFAR-10 and CIFAR-100 | average test accuracy: +2.1% on CIFAR-10 over DivideMix; +5.7% on CIFAR-100 over ELR | Improved |
| Add rotation recognition as an auxiliary task | CIR · CIFAR-10 and CIFAR-100 | average test accuracy: +2.3% on CIFAR-10 over Sel-CL+; +2.3% on CIFAR-100 over Sel-CL+ | Improved |
| Use class-independent multi-binary classifiers instead of a standard softmax classifier | CIFAR noisy-label training · CIFAR-10 and CIFAR-100 | test accuracy: 81.6 73.7 53.5 20.4 90.4 87.0 86.2 82.6 71.9 60.3 38.2 17.1 4.3 71.5 69.6 64.9 53.8 47.5 | Mixed |
| Use MBC-selected clean samples to train SSC (ablation) | CIR · CIFAR-10 Sym-20% | test accuracy: 93.5 -> 95.4 | Improved |
| Use MBC-selected clean samples to train SSC (ablation) | CIR · CIFAR-10 Sym-50% | test accuracy: 90.2 -> 94.2 | Improved |
| Add adaptive thresholding to sample selection (ablation) | CIR · CIFAR-10 Sym-20%, Sym-50%, Asym-20%, Asym-30% | test accuracy: 95.4 -> 95.5; 94.2 -> 94.4; 93.9 -> 94.3; 92.1 -> 93.8 | Improved |
| Add negative class masking, retaining the hardest softmax negative and one random negative (ablation) | CIR · CIFAR-10 Sym-20%, Sym-50%, Asym-20%, Asym-30% | test accuracy: 95.5 -> 95.6; 94.4 -> 95.1; 94.3 -> 94.7; 93.8 -> 94.0 | Improved |
| Use CIR for real-world noisy-label learning | CIR · Clothing1M and Food101N | test accuracy: 74.53 vs 74.37; 87.71 vs 87.50 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DivideMix — DivideMix: Learning with Noisy Labels as Semi-supervised Learning (ICLR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Correlated Input-Dependent Label Noise in Large-Scale Image Classification (CVPR 2021 · similarity 0.88)
- Noise-Robust Learning from Multiple Unsupervised Sources of Inferred Labels (AAAI 2022 · similarity 0.88)
- Foster Adaptivity and Balance in Learning with Noisy Labels (ECCV 2024 · similarity 0.88)
- DISC: Learning From Noisy Labels via Dynamic Instance-Specific Selection and Correction (CVPR 2023 · similarity 0.87)
- Learning from Noisy Labels with No Change to the Training Process (ICML 2021 · similarity 0.87)
- Mitigating Memorization of Noisy Labels via Regularization between Representations (ICLR 2023 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
