Research map / Probabilistic and scientific modeling / Causal discovery
Active Invariant Causal Prediction: Experiment Selection through Stability
Juan L. Gamella, Christina Heinze-Deml
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 |
|---|---|---|---|
| combine Markov and ratio strategies instead of Markov alone (ablation) | A-ICP population policy · Population, 1000 SCMs of size 15 | Jaccard similarity / interventions to exact recovery | Improved |
| add Markov blanket (Lasso) strategy to target selection (ablation) | A-ICP finite-sample policies · Finite regime, 300 SCMs, 10/100/1000 obs per sample | Jaccard similarity at large t | Hurt |
| use empty-set strategy (ablation) | A-ICP finite-sample policy · Finite regime, 300 SCMs | Jaccard similarity / interventions to exact recovery | Improved |
| add ratio and/or Markov strategy on top of empty-set (ablation) | A-ICP empty-set policy · Finite regime, 1000 obs per sample | Jaccard similarity | Mixed |
| use any proposed policy instead of random targets | A-ICP · Population and finite regime | Jaccard similarity | Improved |
| increase observations per interventional sample | A-ICP policies vs random · Finite regime, 10/100/1000 obs per sample | Jaccard similarity gain over random: 10, 100 and 1000 observations per sample | Improved |
| increase intervention strength (shift mean 3, 5, 7, 10) | ratio and empty-set strategies · Finite regime, alpha sweep 0.005-0.1 | improvement over random policy | Improved |
| replace ABCD with A-ICP | ABCD · Finite regime, alpha=0.1 | false positives / family-wise error rate | Improved |
| replace A-ICP with ABCD in early iterations | A-ICP · Finite regime, first few interventions | Jaccard similarity | Hurt |
| increase initial observational sample size (50, 100, 1000) | ABCD · ABCD comparison | Jaccard similarity and FWER: 50, 100 and 1000 | Improved |
| increase initial observational sample size | A-ICP · ABCD comparison | Jaccard similarity | No clear effect |
Closest papers by meaning
- Near-Optimal Experiment Design in Linear non-Gaussian Cyclic Models (NeurIPS 2025 · similarity 0.87)
- Prediction-Intervention Games and Invariant Sets (NeurIPS 2026 · similarity 0.87)
- Bayesian Active Learning for Bivariate Causal Discovery (ICML 2025 · similarity 0.87)
- Causal learning with the invariance principle (NeurIPS 2026 · similarity 0.87)
- ABC3: Active Bayesian Causal Inference with Cohn Criteria in Randomized Experiments (AAAI 2025 · similarity 0.87)
- Data-Driven Covariate Selection for Nonparametric and Cycle-Agnostic Causal Effect Estimation (NeurIPS 2026 · similarity 0.87)
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.
