Research map / Probabilistic and scientific modeling / Causal discovery
Reliable Causal Discovery with Improved Exact Search and Weaker Assumptions
Ignavier Ng, Yujia Zheng, Jiji Zhang, Kun Zhang
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 GLasso instead of MMPC for super-structure (ablation) | A*-SS · exact faithfulness violation SEM, n=20 | simulations missing X-W edge: 69 -> 98 | Improved |
| use GLasso instead of MMPC for super-structure (ablation) | A*-SS · exact faithfulness violation SEM, n=100 | simulations discovering neighbors: 56 -> 100 | Improved |
| use GLasso instead of MMPC for super-structure (ablation) | A*-SS · exact faithfulness violation SEM, n=10^6 | simulations discovering neighbors: 98 -> 100 | Improved |
| use GLasso instead of MMPC (alpha=0.5) for neighbor discovery (ablation) | super-structure estimation · 10-node graphs, degree 5 | TPR: close to 0.7 for MMPC vs close to one for GLasso | Improved |
| GLasso vs MMPC: false discovery rate (ablation) | super-structure estimation · 10-node graphs | FDR | Hurt |
| use GLasso rather than MMPC super-structure for A*-SS (ablation) | A*-SS · 10-node graphs, high degree | CPDAG recovery | Improved |
| add super-structure (A*-SS) and local clusters (Local A*) to A* with known support (ablation) | A* · degree-2 graphs, 15 and 20 nodes, n=10000 | SHD/F1 | Improved |
| Local A* vs A*-SS (ablation) | Local A* · degree-2 graphs, 5-20 nodes | all metrics | No clear effect |
| increase sample size from 300 to 10000 | A*-SS and Local A* · 7-node graphs | advantage over baselines | Hurt |
| add super-structure to A* (ablation) | A*-SS · graph size scalability | max nodes reached: 20 -> 30 nodes | Improved |
| add local search to A*-SS (ablation) | Local A* · graph size scalability | max nodes reached: 30 -> 300 nodes | Improved |
| Local A* vs MMHC | Local A* · large graphs | structure recovery | Improved |
Closest papers by meaning
- Improving Causal Discovery By Optimal Bayesian Network Learning (AAAI 2021 · similarity 0.92)
- BCD Nets: Scalable Variational Approaches for Bayesian Causal Discovery (NeurIPS 2021 · similarity 0.90)
- On Causal Discovery in the Presence of Deterministic Relations (NeurIPS 2024 · similarity 0.89)
- A Meta-Learning Approach to Bayesian Causal Discovery (ICLR 2025 · similarity 0.89)
- Bivariate Causal Discovery using Bayesian Model Selection (ICML 2024 · similarity 0.88)
- THE ROBUSTNESS OF DIFFERENTIABLE CAUSAL DISCOVERY IN MISSPECIFIED SCENARIOS (ICLR 2025 · similarity 0.88)
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.
