Research map / Label-efficient and robust learning / Noisy labels
Mitigating Memorization of Noisy Labels via Regularization between Representations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 representation regularizer to CE | CE on CIFAR10 with symmetric label noise · Symm. CIFAR10, ε = 0.6 | test accuracy (best/last epoch): 61.29/32.83 -> 69.02/65.13 | Helped |
| Add representation regularizer to CE | CE on CIFAR10 with symmetric label noise · Symm. CIFAR10, ε = 0.8 | test accuracy (best/last epoch): 38.46/15.05 -> 61.94/56.78 | Helped |
| Add representation regularizer to GCE | GCE on CIFAR10 with symmetric label noise · Symm. CIFAR10, ε = 0.8 | test accuracy (best/last epoch): 40.71/20.53 -> 63.63/63.05 | Helped |
| Add representation regularizer to Peer Loss | Peer Loss on CIFAR10 with symmetric label noise · Symm. CIFAR10, ε = 0.8 | test accuracy (best/last epoch): 15.60/10.00 -> 61.64/53.52 | Helped |
| Add representation regularizer to Peer Loss | Peer Loss on CIFAR10 with asymmetric label noise · Asymm. CIFAR10, ε = 0.4 | test accuracy (best/last epoch): 84.47/68.93 -> 81.58/75.38 | Mixed |
| Add representation regularizer to CE | CE on CIFAR10N (Worst) · CIFAR10N (Worst) | test accuracy: 77.69 -> 88.74 | Helped |
| Add representation regularizer to CE | CE on CIFAR100N · CIFAR100N | test accuracy: 55.50 -> 60.81 | Helped |
| Add representation regularizer to CE + down-sampling | CE on Clothing1M · Clothing1M | accuracy: 72.95 -> 73.48 | Helped |
| Fix the encoder instead of fine-tuning it | CE with a good pretrained representation · Symmetric label noise, high noise rate | test accuracy | Helped |
| Fix the encoder instead of fine-tuning it | CE with a good pretrained representation · Symmetric label noise, low noise rate | test accuracy | Hurt |
| Down-sample to balance observed noisy-label classes | CE with a fixed encoder · Instance-dependent label noise | test accuracy | Helped |
Closest papers by meaning
- On Learning Contrastive Representations for Learning With Noisy Labels (CVPR 2022 · similarity 0.91)
- Robust early-learning: Hindering the memorization of noisy labels (ICLR 2021 · similarity 0.90)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.89)
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.88)
- $\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise (NeurIPS 2024 · similarity 0.88)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 2025 · similarity 0.88)
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.