Research map / Label-efficient and robust learning / Recognition and long-tail
SelecMix: Debiased Learning by Contradicting-pair Sampling
Inwoo Hwang, Sangjun Lee, Yunhyeok Kwak, Seong Joon Oh, Damien Teney, Jin-Hwa Kim, Byoung-Tak Zhang
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 |
|---|---|---|---|
| Apply SelecMix to LfF | Corrupted CIFAR-10 · Corrupted CIFAR-10, α=0.5% | unbiased accuracy: 30.04 -> 39.44 | Improved |
| Apply SelecMix to LfF | Corrupted CIFAR-10 · Corrupted CIFAR-10, α=1.0% | unbiased accuracy: 33.80 -> 43.68 | Improved |
| Apply SelecMix to vanilla ResNet18 | BFFHQ · BFFHQ, α=0.5% | accuracy on bias-conflicting test examples: 65.60 -> 71.60 | Improved |
| Use both contradicting positives and negatives instead of standard mixup (ablation) | Vanilla model · Corrupted CIFAR-10, α=0.5% | unbiased accuracy: 23.53 -> 38.14 | Improved |
| Use both contradicting pair types rather than positives only (ablation) | Vanilla model · BFFHQ, α=0.5% | unbiased accuracy: 66.20 -> 71.60 | Improved |
| Replace selective pair selection with standard mixup (ablation) | Vanilla model · Colored MNIST, α=1.0% | unbiased accuracy: 50.51 -> 46.19 | Hurt |
| Use the GSC biased embedding rather than the LfF cosine embedding to select contradicting negatives (ablation) | SelecMix · Corrupted CIFAR-10, α=0.5% | unbiased accuracy: 33.69 -> 41.82 | Improved |
| Use the GSC biased embedding rather than the LfF cosine embedding to select contradicting positives (ablation) | SelecMix · Corrupted CIFAR-10, α=0.5% | unbiased accuracy: 40.31 -> 42.94 | Improved |
| Use the GSC biased embedding rather than the LfF cosine embedding to select both pair types (ablation) | SelecMix · Corrupted CIFAR-10, α=0.5% | unbiased accuracy: 38.05 -> 43.68 | Improved |
Closest papers by meaning
- SEBRA : Debiasing through Self-Guided Bias Ranking (ICLR 2025 · similarity 0.88)
- A Simple Remedy for Dataset Bias via Self-Influence: A Mislabeled Sample Perspective (NeurIPS 2024 · similarity 0.85)
- Selective Mixup Helps with Distribution Shifts, But Not (Only) because of Mixup (ICML 2024 · similarity 0.85)
- The Implicit Bias of Structured State Space Models Can Be Poisoned With Clean Labels (NeurIPS 2025 · similarity 0.85)
- Mitigating Spurious Correlations via Disagreement Probability (NeurIPS 2024 · similarity 0.84)
- From Biased Selective Labels to Pseudo-Labels: An Expectation-Maximization Framework for Learning from Biased Decisions (ICML 2024 · similarity 0.84)
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.
