Research map / Label-efficient and robust learning / Recognition and long-tail
Is Importance Weighting Incompatible with Interpolating Classifiers?
Ke Alexander Wang, Niladri Shekhar Chatterji, Saminul Haque, Tatsunori Hashimoto
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 |
|---|---|---|---|
| Replace reweighted cross-entropy with reweighted polynomial-tailed loss | Interpolating neural networks · Imbalanced binary CIFAR10 and subsampled CelebA | test accuracy: as much as 9\% | Improved |
| Exponentiate the importance weights (ablation) | Polynomial-tailed loss · Imbalanced binary CIFAR10 and subsampled CelebA | test accuracy | Improved |
| Exponentiate the importance weights (ablation) | Cross-entropy loss · Imbalanced binary CIFAR10 and subsampled CelebA | test accuracy | No clear effect |
| Use importance weights rather than no importance weighting (ablation) | Polynomial-tailed loss · Imbalanced binary CIFAR10 and subsampled CelebA | test accuracy | Improved |
| Use early stopping (ablation) | Importance-weighted cross-entropy and polynomial-tailed loss · Imbalanced binary CIFAR10 and subsampled CelebA | test accuracy | Improved |
| Use polynomial-tailed loss rather than cross-entropy with early stopping (ablation) | Importance-weighted neural networks · Imbalanced binary CIFAR10 | test accuracy | Improved |
| Use minority-class weight w scaling with τ³ rather than unbiased weight τ (ablation) | Polynomial-tailed linear classifier · Label shift, sub-Gaussian clusters | test error: τ³ | Improved |
| Use importance weights with polynomially-tailed loss rather than exponentially-tailed loss | Overparameterized linear classifier · Label shift, sub-Gaussian clusters | test error: 1/8 | Improved |
| Use polynomial-tailed loss with DRO rather than VS loss with DRO | CelebA classifiers · Subsampled CelebA | test accuracy: 3.5\% | Improved |
| Use polynomial-tailed loss with DRO rather than VS loss with DRO | CelebA classifiers · Subsampled CelebA | worst group accuracy: 8\% | Improved |
Closest papers by meaning
- Optimizing importance weighting in the presence of sub-population shifts (ICLR 2025 · similarity 0.90)
- How Uniform Random Weights Induce Non-uniform Bias: Typical Interpolating Neural Networks Generalize with Narrow Teachers (ICML 2024 · similarity 0.85)
- Thumb on the Scale: Optimal Loss Weighting in Last Layer Retraining (NeurIPS 2025 · similarity 0.85)
- Noisy Interpolation Learning with Shallow Univariate ReLU Networks (ICLR 2024 · similarity 0.84)
- Rethinking Benign Overfitting in Two-Layer Neural Networks (ICML 2025 · similarity 0.84)
- Bias of Stochastic Gradient Descent or the Architecture: Disentangling the Effects of Overparameterization of Neural Networks (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.
