Research map / Label-efficient and robust learning / Noisy labels
Joint Asymmetric Loss for Learning with Noisy Labels
Jialiang Wang, Xianming Liu, Xiong Zhou, Gangfeng Hu, Deming Zhai, Junjun Jiang, Xiangyang Ji
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 |
|---|---|---|---|
| add AMSE passive loss to NCE (JAL-CE vs NCE alone) (ablation) | JAL-CE · CIFAR-10 symmetric noise 0.8 | test accuracy (%): 41.20 -> 65.43 | Improved |
| add active NCE to AMSE (JAL-CE vs AMSE alone) (ablation) | JAL-CE · CIFAR-10 asymmetric noise 0.4 | test accuracy (%): 58.07 -> 79.54 | Improved |
| add active NCE to AMSE under symmetric noise (ablation) | JAL-CE · CIFAR-10 symmetric noise 0.8 | test accuracy (%): 64.97 -> 65.43 | No clear effect |
| add AMSE to NCE under asymmetric noise 0.4 (ablation) | JAL-CE · CIFAR-10 asymmetric noise 0.4 | test accuracy (%): 65.33 -> 79.54 | Improved |
| replace NCE+RCE with asymmetric AGCE+MAE in APL framework (ablation) | APL · CIFAR-10 asymmetric noise 0.4 | test accuracy (%): 77.94 -> 57.80 | Hurt |
| increase alpha and beta from 1 to 10 in AGCE+MAE (ablation) | AGCE+MAE · CIFAR-10 symmetric noise 0.8 | test accuracy (%): 44.61 -> 21.83 | Hurt |
| larger AMSE parameter a (20, 30, 40 vs 10) (ablation) | AMSE · CIFAR-10 symmetric noise 0.8 | test accuracy (%) | Improved |
| too large a in AMSE (ablation) | AMSE · CIFAR-10 asymmetric noise | test accuracy (%) | Hurt |
| use q=3 instead of q=2 in AMSE (ablation) | AMSE · CIFAR-10 symmetric noise 0.8 | test accuracy (%) | No clear effect |
| use q=1 instead of q=2 in AMSE (ablation) | AMSE · CIFAR-10 symmetric noise 0.8 | test accuracy (%) | Hurt |
| replace ANL passive NNCE with AMSE (JAL-CE vs ANL-CE) | JAL-CE · CIFAR-10 symmetric noise 0.8 | test accuracy (%): 62.28 -> 65.43 | Improved |
| JAL-CE vs ANL-FL under high symmetric noise with 100 classes | JAL-CE · CIFAR-100 symmetric noise 0.8 | test accuracy (%): 28.81 -> 22.80 | Hurt |
Closest papers by meaning
- Asymmetric Loss Functions for Learning with Noisy Labels (ICML 2021 · similarity 0.94)
- Normalized Loss Functions for Deep Learning with Noisy Labels (ICML 2020 · similarity 0.92)
- Asymmetric Loss Functions for Noise-Tolerant Learning: Theory and Applications (TPAMI 2023 · similarity 0.91)
- $\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise (NeurIPS 2024 · similarity 0.89)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · similarity 0.89)
- Regroup Median Loss for Combating Label Noise (AAAI 2024 · 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.
