Research map / Label-efficient and robust learning / Noisy labels
UniCon: Combating Label Noise Through Uniform Selection and Contrastive Learning
Check 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 |
|---|---|---|---|
| Use class-balanced uniform clean-sample selection (ablation) | UniCon · CIFAR10 and CIFAR100, 90% symmetric noise | best test accuracy: 85.49 -> 90.81; 39.87 -> 44.82 | Helped |
| Add contrastive learning (ablation) | UniCon · CIFAR10 and CIFAR100, 90% symmetric noise | best test accuracy: 87.28 -> 90.81; 41.83 -> 44.82 | Helped |
| Ensemble both networks' predictions during sample separation (ablation) | UniCon · CIFAR10 and CIFAR100, 90% symmetric noise | best test accuracy: 88.84 -> 90.81; 42.36 -> 44.82 | Helped |
| Add the unsupervised loss L_U (ablation) | UniCon · CIFAR10 and CIFAR100, 80% symmetric noise | best test accuracy: 87.82 -> 93.97; 56.94 -> 63.98 | Helped |
| Add the regularization loss L_reg (ablation) | UniCon · CIFAR10 and CIFAR100, 80% symmetric noise | best test accuracy: 93.59 -> 93.97; 61.75 -> 63.98 | Helped |
| Use UniCon rather than the strongest reported baseline | CIFAR10 · 20% symmetric noise | test accuracy: 96.1 vs 96.0 | No clear effect |
| Use UniCon rather than the strongest reported baseline | CIFAR10 · 10% asymmetric noise | test accuracy: 95.4 vs 95.3 | No clear effect |
| Use UniCon rather than MOIT | WebVision · WebVision validation | Top-1 accuracy: 78.76 -> 77.60 | Hurt |
| Use UniCon rather than the strongest baseline | Tiny-ImageNet · symmetric label noise | best test accuracy: +1% performance improvement over SOTA for all noise rates | Helped |
Closest papers by meaning
- Readiness-Aware Sample Selection for Noisy Labels with Class Imbalance (NeurIPS 2026 · similarity 0.88)
- Tackling Instance-Dependent Label Noise via a Universal Probabilistic Model (AAAI 2021 · similarity 0.88)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.87)
- Enhancing Sample Selection Against Label Noise by Cutting Mislabeled Easy Examples (NeurIPS 2025 · similarity 0.87)
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.86)
- Addressing Multi-Label Learning with Partial Labels: From Sample Selection to Label Selection (AAAI 2025 · similarity 0.86)
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.