Research map / Label-efficient and robust learning / Domain adaptation and generalization
Environment Inference for Invariant Learning
Elliot Creager, Joern-Henrik Jacobsen, Richard Zemel
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 |
|---|---|---|---|
| Infer environments from aggregated data and use IRM instead of handcrafted environments | CMNIST invariant learning · CMNIST test | accuracy (%): 65.5 ± 2.3 -> 68.4 ± 2.7 | Improved |
| Infer environments and use GroupDRO rather than ERM | Waterbirds · Waterbirds test (worst group) | accuracy (%): 60.3 -> 78.7 | Improved |
| Infer environments and use GroupDRO rather than ERM | Waterbirds · Waterbirds test (avg) | accuracy (%): 97.3 -> 96.9 | Hurt |
| Infer environments and apply invariant learning rather than ERM | Adult-Confounded · Adult-Confounded test | accuracy: 31.1 ± 4.4 -> 78.8 ± 1.4 | Improved |
| Infer environments using binning and apply GroupDRO rather than ERM | CivilComments-WILDS · CivilComments-WILDS test (worst group) | accuracy: 61.6 ± 1.3 -> 67.0 ± 2.4 | Improved |
| Infer environments using binning and apply GroupDRO rather than ERM | CivilComments-WILDS · CivilComments-WILDS test (avg) | accuracy: 92.0 ± 0.4 -> 90.5 ± 0.2 | Hurt |
| Use an ERM reference that learns the spurious color feature rather than a reference that does not expose the relevant bias | EIIL on CMNIST · CMNIST | generalization | Mixed |
| Use a hand-coded color-based predictor as the reference model | EIIL on CMNIST · CMNIST | generalization | Improved |
| Jointly train the classifier and environment assignments with alternating updates instead of the sequential two-stage procedure (ablation) | EIIL · Experiments reported in the paper | performance | No clear effect |
Closest papers by meaning
- Provably Invariant Learning without Domain Information (ICML 2023 · similarity 0.88)
- Environment Diversification with Multi-head Neural Network for Invariant Learning (NeurIPS 2022 · similarity 0.87)
- Free Lunch for Domain Adversarial Training: Environment Label Smoothing (ICLR 2023 · similarity 0.86)
- Sufficient Invariant Learning for Distribution Shift (CVPR 2025 · similarity 0.86)
- Enhancing Graph Invariant Learning from a Negative Inference Perspective (ICML 2025 · similarity 0.86)
- Learning Conditional Invariances through Non-Commutativity (ICLR 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.
