Research map / Label-efficient and robust learning / Recognition and long-tail
Causal Effect Regularization: Automated Detection and Removal of Spurious Correlations
Abhinav Kumar, Amit Deshpande, Amit Sharma
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 Riesz estimator instead of Direct for causal effect (ablation) | AutoACER effect estimation · Syn-Text-Obs-Conf, kappa 0.99 | estimated effect (true 0): 0.22 -> 0.15 | Improved |
| use Riesz instead of Direct estimator (ablation) | AutoACER effect estimation · MNIST34, kappa 0.8 | estimated effect (true 0): 0.06 -> 0.04 | Improved |
| use Riesz instead of Direct estimator when effect not identified (ablation) | AutoACER effect estimation · Syn-Text-Unobs-Conf, kappa 0.99 | estimated effect (true 0): 0.67 -> 0.70 | Hurt |
| use Riesz instead of Direct estimator (ablation) | AutoACER effect estimation · MNIST34, kappa 0.99 | estimated effect (true 0): 0.27 -> 0.31 | Hurt |
| raise spurious predictive correlation kappa from 0.5 to 0.99 | causal effect estimators · Syn-Text-Unobs-Conf, Riesz | estimated effect (true 0): 0.06 -> 0.70 | Hurt |
| soft causal-effect regularization (AutoACER) instead of hard Mouli+CAD detection and augmentation | AutoACER · Syn-Text-Unobs-Conf | average group accuracy and DeltaProb | Improved |
| AutoACER vs baselines | AutoACER · MNIST34 | DeltaProb | Improved |
| AutoACER vs ERM on accuracy | AutoACER · Twitter-AAE | average group accuracy | No clear effect |
| Mouli+CAD invariance via counterfactual augmentation | Mouli+CAD · Twitter-AAE | DeltaProb | Hurt |
| Mouli+CAD at high correlation (kappa >= 0.8) | Mouli+CAD · Syn-Text-Unobs-Conf | spurious attribute detection | Hurt |
| replace CAD with causal effect regularization in Mouli pipeline (Mouli+AutoACER) (ablation) | Mouli · Syn-Text, MNIST34, Twitter-AAE | invariance to spurious attribute | Improved |
| give all methods knowledge of the spurious attribute (ablation) | AutoACER and baselines · all datasets (Fig. 7) | average group accuracy and DeltaProb | No clear effect |
Closest papers by meaning
- Robustness to Spurious Correlations in Text Classification via Automatically Generated Counterfactuals (AAAI 2021 · similarity 0.87)
- Let Samples Speak: Mitigating Spurious Correlation by Exploiting the Clusterness of Samples (CVPR 2025 · similarity 0.85)
- An Analysis of Causal Effect Estimation using Outcome Invariant Data Augmentation (NeurIPS 2025 · similarity 0.85)
- Spurious Correlations in High Dimensional Regression: The Roles of Regularization, Simplicity Bias and Over-Parameterization (ICML 2025 · similarity 0.85)
- Mitigating Spurious Correlations via Disagreement Probability (NeurIPS 2024 · similarity 0.84)
- Do causal predictors generalize better to new domains? (NeurIPS 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.
