Research map / Label-efficient and robust learning / Noisy labels
DISC: Learning From Noisy Labels via Dynamic Instance-Specific Selection and Correction
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 |
|---|---|---|---|
| Use dynamic instance-specific thresholds and two-view selection/correction | DISC · CIFAR-100, IDN 60% | accuracy: 59.40 ± 0.46 -> 69.57 ± 0.14 | Helped |
| Use DISC instead of the strongest reported baseline | Tiny ImageNet · symmetric noise 50% | best test accuracy: 52.7 -> 64.3 | Helped |
| Use DISC instead of the strongest reported baseline | Animals-10N · Animals-10N | accuracy: 84.2 ± 0.07 -> 87.1 ± 0.15 | Helped |
| Use DISC instead of the strongest reported baseline | Food-101 · Food-101 | accuracy: 87.57 -> 89.02 | Helped |
| Replace GMM division with DIST in DivideMix | DivideMix · Clothing1M | accuracy: 74.45 -> 74.79 | Helped |
| Use weak and strong augmented views instead of one weak view (ablation) | DISC · CIFAR-10, IDN 40% | accuracy: 94.32 -> 95.94 | Helped |
| Use weak and strong augmented views instead of one weak view (ablation) | DISC · CIFAR-100, IDN 40% | accuracy: 75.21 -> 78.44 | Helped |
| Use two weakly augmented views instead of one weak view (ablation) | DISC · CIFAR-10, IDN 20% | accuracy: 95.73 -> 96.20 | Helped |
| Use weak and strong views instead of two weak views (ablation) | DISC · CIFAR-100, IDN 20% | accuracy: 79.24 -> 80.12 | Helped |
| Use dynamic instance-specific thresholding rather than fixed threshold 0.5 (ablation) | Instance selection for noisy-label learning · CIFAR-100, IDN 60% | accuracy: 14.78 -> 60.07 | Helped |
| Use dynamic instance-specific thresholding rather than GMM selection (ablation) | Instance selection for noisy-label learning · CIFAR-100, IDN 40% | accuracy: 30.73 -> 70.11 | Helped |
| Add a second view to the one-view DISC-W variant (ablation) | DISC · CIFAR-10, IDN 60% | accuracy: 88.37 -> 93.65 | Helped |
Closest papers by meaning
- Tackling Instance-Dependent Label Noise via a Universal Probabilistic Model (AAAI 2021 · similarity 0.90)
- Noise-Robust Learning from Multiple Unsupervised Sources of Inferred Labels (AAAI 2022 · similarity 0.90)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.89)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.88)
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.88)
- Revisiting Interpolation for Noisy Label Correction (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.