Research map / Label-efficient and robust learning / Noisy labels
Denoising after Entropy-Based Debiasing a Robust Training Method for Dataset Bias with Noisy Labels
Sumyeong Ahn, Se-Young Yun
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 |
|---|---|---|---|
| Use DENEB instead of Vanilla | Simple-ConvNet · Colored MNIST (α=1%, η=10%) | unbiased test accuracy: 39.24 { $\pm$ 1.91}\% -> 91.81 { $\pm$ 0.84}\% | Improved |
| Use DENEB instead of Vanilla | Simple-ConvNet · Colored MNIST (α=5%, η=50%) | unbiased test accuracy: 70.13\% { $\pm$ 3.42} \% -> 94.55\% { $\pm$ 0.22} \% | Improved |
| Use DENEB instead of Vanilla | ResNet-18 · Corrupted CIFAR-10 (α=1%, η=10%) | unbiased test accuracy: 25.43\% { $\pm$ 0.84} \% -> 26.05\% { $\pm$ 0.54} \% | No clear effect |
| Use DENEB instead of Vanilla | ResNet-18 · Corrupted CIFAR-10 (α=5%, η=50%) | unbiased test accuracy: 31.86\% { $\pm$ 0.96} \% -> 35.32\% { $\pm$ 1.03} \% | Improved |
| Use DENEB instead of Vanilla | ResNet-18 · BAR (η=10%) | unbiased test accuracy: 54.37 { $\pm$ 1.10}\% -> 62.30 { $\pm$ 0.91}\% | Improved |
| Use DENEB instead of Vanilla | ResNet-18 · BFFHQ (η=10%) | unbiased test accuracy: 71.38 { $\pm$ 0.58}\% -> 75.24 { $\pm$ 0.68}\% | Improved |
| Run denoising before debiasing (ablation) | DBwNL training · DBwNL benchmarks | retention of bias-conflicting samples | Hurt |
| Emphasize bias-conflicting samples before denoising (ablation) | DBwNL training · DBwNL denoising analysis | retention of bias-conflicting samples | Improved |
| Use entropy scores rather than given-label-based difficulty scores (ablation) | Debiasing on DBwNL · Colored MNIST and waterbird analysis | unbiased test accuracy | Improved |
| Apply a 50× loss weight to bias-conflicting samples during denoising (ablation) | Denoising algorithms on DBwNL · DBwNL denoising analysis | retention of bias-conflicting samples: 50× | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DivideMix — DivideMix: Learning with Noisy Labels as Semi-supervised Learning (ICLR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Simple Remedy for Dataset Bias via Self-Influence: A Mislabeled Sample Perspective (NeurIPS 2024 · similarity 0.86)
- Diffusing DeBias: Synthetic Bias Amplification for Model Debiasing (NeurIPS 2025 · similarity 0.86)
- Mitigating Spurious Correlations via Disagreement Probability (NeurIPS 2024 · similarity 0.86)
- Ameliorate Spurious Correlations in Dataset Condensation (ICML 2024 · similarity 0.85)
- Self-supervised Debiasing Using Low Rank Regularization (CVPR 2024 · similarity 0.85)
- DeNetDM: Debiasing by Network Depth Modulation (NeurIPS 2024 · similarity 0.85)
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.
