Research map / Learning theory and algorithms / Prediction and decision losses
What Is Missing in IRM Training and Evaluation? Challenges and Solutions
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 small-batch rather than original large-batch training (ablation) | IRM methods · CMNIST and CFMNIST | Avg Acc: 1% ∼ 3% improvement | Helped |
| Use small-batch training rather than original large-batch training (ablation) | IRM · CMNIST | Avg Acc: 67.13 -> 68.33 | Helped |
| Use small-batch training rather than original large-batch training (ablation) | IRM · CFMNIST | Avg Acc: 67.19 -> 68.76 | Helped |
| Use small-batch rather than original large-batch training (ablation) | IRM methods · CMNIST and CFMNIST | Acc Gap: 3.43 -> 2.04; 3.35 -> 1.45 | Helped |
| Replace one test environment with diverse test-time environments | IRM evaluation · CMNIST | method ranking across environments | Mixed |
| Add a consensus prediction head and per-environment stationarity regularization | IRM-Game · CMNIST | Avg Acc: 65.69 -> 69.47 | Helped |
| Add per-environment stationarity regularization to consensus optimization (ablation) | BLOC-IRM · CMNIST | Avg Acc: 67.14 -> 69.47 | Helped |
| Replace stationarity regularization with REx regularization (ablation) | BLOC-IRM · CMNIST | Avg Acc: 69.47 -> 62.71 | Hurt |
| Replace stationarity regularization with Fishr regularization (ablation) | BLOC-IRM · CMNIST | Avg Acc: 69.47 -> 63.25 | Hurt |
| Use BLOC-IRM rather than the strongest baseline | IRM methods · PACS | Avg Acc: 70.21 -> 70.31 | Helped |
| Increase model size beyond d = 390 (ablation) | IRM methods · CMNIST | Avg Acc | Hurt |
| Train on additional covariate shifts (ablation) | IRM methods · CMNIST and CFMNIST | Avg Acc | Hurt |
Closest papers by meaning
- Invariant Risk Minimization Is A Total Variation Model (ICML 2024 · similarity 0.89)
- Out-of-distribution Generalization for Total Variation based Invariant Risk Minimization (ICLR 2025 · similarity 0.84)
- Data Mixture Optimization: A Multi-fidelity Multi-scale Bayesian Framework (NeurIPS 2025 · similarity 0.83)
- The Implicit Bias of Heterogeneity towards Invariance: A Study of Multi-Environment Matrix Sensing (NeurIPS 2024 · similarity 0.83)
- Boosting Test Performance with Importance Sampling--a Subpopulation Perspective (AAAI 2025 · similarity 0.83)
- Structure-informed Risk Minimization for Robust Ensemble Learning (ICML 2025 · similarity 0.82)
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.