Research map / Label-efficient and robust learning / Test-time adaptation
Statistical Learning under Heterogenous Distribution Shift
Max Simchowitz, Anurag Ajay, Pulkit Agrawal, Akshay Krishnamurthy
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 additive predictor f(x)+g(y) fit by ERM (structure separating simple feature x from complex feature y) | ERM under heterogeneous covariate shift · Synthetic additive regression | MSE under distribution shift | Improved |
| Shift in simpler feature (background) instead of complex feature (birdtype) | Standard-trained classifier on Waterbird · Waterbird | birdtype prediction accuracy | No clear effect |
| Shift in simpler attribute (bald) instead of complex attribute (big_lips) | Logical operator classifiers on CelebA · CelebA OR/XOR | success rate under shift | Improved |
| Shift in simpler factor (object mass) instead of complex factor (dampness) | Imitation policy on Pusher Control · Pusher Control | success rate under shift | Improved |
| Replace naive covariate-shift bound with additive decomposition bound (nu_1 Rtrain[f] + nu_2 Rtrain(f,g)) | ERM theory · Theoretical bound | excess test risk bound | Improved |
| Holder-style Dudley inequality for Hadamard products used to bound cross-critical radius (vs multiplying Dudley functionals) | Cross-critical radius bound · Theory | bound on cross-critical radius | Improved |
| Replace orthogonal ML double-training with single-step ERM without auxiliary supervision | Estimation of f_star · Theory vs orthogonal ML | rate for f recovery | Mixed |
Closest papers by meaning
- Distribution Shift Is Key to Learning Invariant Prediction (AAAI 2026 · similarity 0.87)
- Overparameterization Improves Robustness to Covariate Shift in High Dimensions (NeurIPS 2021 · similarity 0.87)
- Tolerant Algorithms for Learning with Arbitrary Covariate Shift (NeurIPS 2024 · similarity 0.85)
- On the Asymptotic Distribution of the Minimum Empirical Risk (ICML 2024 · similarity 0.85)
- Not all distributional shifts are equal: Fine-grained robust conformal inference (ICML 2024 · similarity 0.85)
- Generalization vs Specialization under Concept Shift (NeurIPS 2025 · 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.
