Research map / Label-efficient and robust learning / Noisy labels
Instance-dependent Label Distribution Estimation for Learning with Label Noise
Zehui Liao, Shishuai Hu, Yutong Xie, Yong Xia
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 the LD supervision loss L_LD to cross-entropy (ablation) | ILDE · CIFAR-10 (asymmetric noise 0.5) | test accuracy (%): 46.81 -> 70.19 | Improved |
| add the diagonally-dominant regularizer L_M (ablation) | ILDE · CIFAR-10 (asymmetric noise 0.5) | test accuracy (%): 70.19 -> 80.74 | Improved |
| add temporal ensembling of the estimated LD (ablation) | ILDE · CIFAR-10 (asymmetric noise 0.5) | test accuracy (%): 75.85 -> 79.71 | Improved |
| add sharpening of the estimated LD (ablation) | ILDE · CIFAR-10 (asymmetric noise 0.5) | test accuracy (%): 79.71 -> 80.74 | Improved |
| reduce batch size from 128 to 32 (ablation) | ILDE · CIFAR-10 (asymmetric noise 0.5) | test accuracy (%): 80.74 -> 84.43 | Improved |
| reduce batch size further from 32 to 8 (ablation) | ILDE · CIFAR-10 (asymmetric noise 0.5) | test accuracy (%): 84.43 -> 78.83 | Hurt |
| tune hyperparameters per noise rate (Ours* vs Ours) (ablation) | ILDE · CIFAR-10 (symmetric noise 0.8) | test accuracy (%): 56.25 -> 59.10 | Improved |
| use ILDE instead of MEIDTM | ILDE* · CIFAR-10 (symmetric noise 0.8) | test accuracy (%): 55.84 -> 59.10 | Improved |
| use ILDE instead of MEIDTM | ILDE* · CIFAR-10 (asymmetric noise 0.5) | test accuracy (%): 77.23 -> 80.74 | Improved |
| use ILDE instead of best competitor | ILDE · CIFAR-100N | test accuracy (%): 55.06 -> 55.95 | Improved |
| use ILDE instead of MEIDTM on saturated MNIST | ILDE* · MNIST (symmetric noise 0.8) | test accuracy (%): 98.35 -> 98.57 | No clear effect |
| use ILDE instead of TNLPAD | ILDE · Clothing1M | test accuracy (%): 71.51 -> 72.45 | Improved |
Closest papers by meaning
- A Parametrical Model for Instance-Dependent Label Noise (TPAMI 2023 · similarity 0.92)
- Extended : Learning with Mixed Closed-set and Open-set Noisy Labels (TPAMI 2022 · similarity 0.91)
- Estimation of the Label-Noise Transition Matrix with Performance Guarantees via Selective Classification (NeurIPS 2026 · similarity 0.89)
- Learning the Latent Causal Structure for Modeling Label Noise (NeurIPS 2024 · similarity 0.89)
- Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning (CVPR 2024 · similarity 0.89)
- Learning Causal Transition Matrix for Instance-dependent Label Noise (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.
