Research map / Label-efficient and robust learning / Noisy labels
Safeguarded Dynamic Label Regression for Noisy Supervision
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 |
|---|---|---|---|
| Replace Softmax noise-transition layer (S-adaptation) with Bayesian Dirichlet LCCN and dynamic label regression | noisy-label classifier · CIFAR-10 r=0.7 | accuracy: 72.74 -> 79.48 | Helped |
| Replace S-adaptation with LCCN | noisy-label classifier · CIFAR-100 r=0.4 | accuracy: 55.27 -> 65.52 (vs best baseline) | Helped |
| Replace baselines with LCCN at extreme symmetric-like noise | noisy-label classifier · CIFAR-100 r=0.5 | accuracy: 33.71 vs 34.52 (Bootstrapping) | No clear effect |
| Replace S-adaptation with LCCN on real noisy data | noisy-label classifier · Clothing1M | accuracy: 70.36 -> 73.07 | Helped |
| Replace baselines with LCCN on WebVision | noisy-label classifier · WebVision | top-1 accuracy: 63.20 -> 63.52 | Helped |
| Remove manually estimated transition warm-up (ablation) | LCCN · Clothing1M | accuracy: 73.07 -> 71.63 | Hurt |
| Add slight tempering of classifier prediction in sampling (without warm-up) (ablation) | LCCN · Clothing1M | accuracy: 71.63% -> 72.92% | Helped |
| Bayesian bounded transition update vs Softmax-layer transition update | LCCN vs S-adaptation · CIFAR-10 r=0.5 | stability of test accuracy | Helped |
Closest papers by meaning
- Beyond Class-Conditional Assumption: A Primary Attempt to Combat Instance-Dependent Label Noise (AAAI 2021 · similarity 0.91)
- Learning from Noisy Labels with No Change to the Training Process (ICML 2021 · similarity 0.90)
- Handling Label Noise via Instance-Level Difficulty Modeling and Dynamic Optimization (NeurIPS 2025 · similarity 0.90)
- Learning the Latent Causal Structure for Modeling Label Noise (NeurIPS 2024 · similarity 0.89)
- Learning from Noisy Labels via Conditional Distributionally Robust Optimization (NeurIPS 2024 · similarity 0.88)
- Label Noise: Ignorance Is Bliss (NeurIPS 2024 · 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.