Research map / Label-efficient and robust learning / Domain adaptation and generalization
Learning Invariant Representations with a Nonparametric Nadaraya-Watson Head
Alan Q. Wang, Minh Nguyen, Mert R. Sabuncu
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 environment-aware support set (NW^{B,e} Full vs ERM) | NW head · Camelyon-17 | average accuracy: 70.3 -> 80.0 | Improved |
| add environment-aware support set (NW^{B,e} Full vs ERM) | NW head · ISIC | F1-score: 58.2 -> 69.6 | Improved |
| use NW^{B,e} Ensemble instead of the best baseline CLoVE | NW head · FMoW | worst-region accuracy: 40.1 -> 37.8 | Hurt |
| use Cluster inference instead of Full for NW^{B,e} (ablation) | NW^{B,e} · ISIC | F1-score: 69.6 -> 71.1 | Improved |
| class-balanced support set (NW^B) vs ERM (ablation) | NW^B · ISIC | F1-score: 58.2 -> 61.9 | Improved |
| class-balanced support set (NW^B) vs ERM (ablation) | NW^B · Camelyon-17 | average accuracy: 70.3 -> 72.0 | No clear effect |
| add environment-aware support (NW^{B,e}) on top of NW^B (ablation) | NW head · ISIC | F1-score: 61.9 -> 69.6 | Improved |
| ERM with balanced classes per environment instead of NW^B (ablation) | ERM^B · ISIC | F1-score: 63.0 | No clear effect |
| explicit invariance regularizer instead of implicit training (ablation) | NW^{B,e} Full · Camelyon-17 | average accuracy: 80.0 -> 79.7 | No clear effect |
| Probe inference instead of Full for NW^{B,e} (ablation) | NW^{B,e} · Camelyon-17 | average accuracy: 80.0 -> 75.3 | Hurt |
| Random support inference instead of Full for NW^{B,e} (ablation) | NW^{B,e} · FMoW | worst-region accuracy: 35.0 -> 31.2 | Hurt |
| NW^{B,e} Probe vs IRM | NW^{B,e} Probe · Camelyon-17 | average accuracy: 70.9 -> 75.3 | Improved |
Closest papers by meaning
- Invariant Causal Representation Learning for Out-of-Distribution Generalization (ICLR 2022 · similarity 0.87)
- Invariant Representations without Adversarial Training (NeurIPS 2018 · similarity 0.86)
- Invariant-Equivariant Representation Learning for Multi-Class Data (ICML 2019 · similarity 0.86)
- Fundamental Limits and Tradeoffs in Invariant Representation Learning (NeurIPS 2023 · similarity 0.86)
- Unsupervised Representation Learning - an Invariant Risk Minimization Perspective (ICLR 2026 · similarity 0.86)
- Unsupervised Representation Learning by Invariance Propagation (NeurIPS 2020 · similarity 0.86)
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.
