Research map / Probabilistic and scientific modeling / Causal discovery
Near-Optimal Multi-Perturbation Experimental Design for Causal Structure Learning
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 DGC with multi-perturbation interventions (q=3) instead of random interventions or greedy single-perturbation interventions | Edge-orientation on ER(0.1) graphs, p=40 · Synthetic ER(0.1) graphs | proportion of edges identified | Helped |
| Use graph-dependent SSG-b rather than graph-agnostic SSG-a | Edge-orientation on synthetic graphs · Synthetic random graphs | proportion of edges identified | Helped |
| Increase the intervention size q for SSG-b | Edge-orientation on a fully connected graph (p=5, m=2) · Fully connected graph | proportion of edges identified | No clear effect |
| Use DGC or SSG-b rather than SSG-a | Edge-orientation on three disconnected star graphs · Synthetic forest of 3 disconnected star graphs | proportion of edges identified | Helped |
| Use multi-perturbation methods (q=5) rather than greedy single-perturbation interventions | Finite-sample causal structure learning on random linear SEMs · Synthetic finite-sample graphs | mutual information objective | Helped |
| Optimize the finite-sample mutual-information objective directly rather than its infinite-sample approximation | SSG-b · Synthetic finite-sample graphs | mutual information objective | Helped |
| Optimize edge orientation rather than mutual information for the finite-sample objective | DGC-infinity · Synthetic finite-sample graphs | mutual information objective | Hurt |
| Use multi-perturbation methods (q=5) rather than random or greedy single-perturbation interventions | DREAM3 Yeast1 network · DREAM3 Yeast1 | proportion of unknown edges oriented | Helped |
| Use multi-perturbation interventions rather than single-perturbation interventions | Causal structure learning under infinite samples · Synthetic random graphs | proportion of edges identified | Helped |
Closest papers by meaning
- Budgeted Experiment Design for Causal Structure Learning (ICML 2018 · similarity 0.93)
- Differentiable Multi-Target Causal Bayesian Experimental Design (ICML 2023 · similarity 0.92)
- Near-Optimal Experiment Design in Linear non-Gaussian Cyclic Models (NeurIPS 2025 · similarity 0.90)
- Causal Discovery from Soft Interventions with Unknown Targets: Characterization and Learning (NeurIPS 2020 · similarity 0.90)
- Optimal Experiments for Partial Causal Effect Identification (NeurIPS 2026 · similarity 0.87)
- Partial Structure Discovery is Sufficient for No-regret Learning in Causal Bandits (NeurIPS 2024 · 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.