Research map / Graph learning / Graph algorithms and clustering
On Learning Necessary and Sufficient Causal Graphs
Hengrui Cai, Yixin Wang, Michael Jordan, Rui Song
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 total effect instead of direct effect as criterion (ablation) | NSCSL · S2 p=5, n=100 | TPR: 0.50 -> 0.99 | Improved |
| use total effect instead of direct effect as criterion (ablation) | NSCSL · S4 ER p=20, n=1000 | TPR: 0.67 -> 1.00 | Improved |
| use total effect instead of direct effect as criterion (ablation) | NSCSL · S5 ER p=50, n=3000 | SHD: 9.76 -> 1.58 | Improved |
| use total effect instead of direct effect as criterion (ablation) | NSCSL · S1 p=5, n=100 | SHD: 0.14 -> 0.14 | No clear effect |
| add necessary-and-sufficient causal effect penalty to NOTEARS | NOTEARS · S1 p=5, n=100 | SHD: 2.02 -> 0.14 | Improved |
| add necessary-and-sufficient causal effect penalty to NOTEARS | NOTEARS · S4 ER p=20, n=1000 | SHD: 36.40 -> 0.04 | Improved |
| add causal effect penalty to NOTEARS in a setting with no spurious variables | NOTEARS · S3 p=5, n=100 | SHD: 0.06 -> 0.06 | No clear effect |
| add causal effect penalty to NOTEARS under nonlinear SEM | NOTEARS · S4 nonlinear p=20, n=1000 | SHD: 35.90 -> 0.60 | Improved |
| add causal effect penalty to NOTEARS under scale-free graph | NOTEARS · S5 SF p=50, n=1000 | SHD: 123.10 -> 5.08 | Improved |
| increase sample size from 30 to 100 | NSCSL-TE · S1 p=5 | SHD: 0.64 -> 0.14 | Improved |
| increase sample size from 100 to 1000 | NSCSL-TE · S4 ER p=20 | runtime: 10.8 -> 56.2 | Hurt |
| add causal effect penalty to NOTEARS on single-cell protein data | NOTEARS · Akt protein network | SHD: 21 -> 8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DAGGNN — DAG-GNN: DAG Structure Learning with Graph Neural Networks (ICML 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Local Identifying Causal Relations in the Presence of Latent Variables (ICML 2025 · similarity 0.90)
- Rethinking Causal Relationships Learning in Graph Neural Networks (AAAI 2024 · similarity 0.88)
- Learning Mixtures of Unknown Causal Interventions (NeurIPS 2024 · similarity 0.88)
- A Recipe for Causal Graph Regression: Confounding Effects Revisited (ICML 2025 · similarity 0.88)
- CausalTab: Pretraining Across Causal Environments for Tabular Causal Discovery (NeurIPS 2026 · similarity 0.88)
- Characterization and Learning of Causal Graphs from Hard Interventions (NeurIPS 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.
