Research map / Probabilistic and scientific modeling / Causal discovery
Embracing Discrete Search: A Reasonable Approach to Causal Structure Learning
Marcel Wienöbst, Leonard Henckel, Sebastian Weichwald
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 |
|---|---|---|---|
| add iterated local search restarts (20 to 100) to FLOP (ablation) | FLOP · Alarm network | % target CPDAG recovered: 2% -> 74% (20 restarts), 82% (100 restarts) | Improved |
| add iterated local search restarts to FLOP (ablation) | FLOP · SF graphs | % instances with BIC at least as good as true graph: 6% -> 76% (20), 94% (100) | Improved |
| add iterated local search restarts to FLOP (ablation) | FLOP · ER graphs (25 nodes, degree 16, 50,000 samples) | % target graph recovered: 26% (20) -> 50% (100) -> 56% (500) | Improved |
| order-based discrete search (FLOP) instead of BOSS | FLOP · ER graphs (BIC) | % BIC at least as good as true graph: 52% (BOSS) -> 84% | Improved |
| more ILS restarts when BIC is misspecified (Pathfinder) (ablation) | FLOP · Pathfinder network | SHD | Hurt |
| run FLOP with no ILS perturbations (x=0) (ablation) | FLOP · ER graphs, perturbation sweep | SHD: SHD often in the hundreds without ILS | Hurt |
| raise BIC penalty lambda_BIC to at least 2 (ablation) | FLOP_20 · synthetic graphs | SHD | Improved |
| FLOP vs DAGMA on Sachs | FLOP_20 · Sachs | average SHD: 12.58 vs 11.7 (DAGMA) | No clear effect |
| Cholesky-based order initialization with FLOP_0 vs PC/GES on paths | FLOP_0 · 50-node paths | % target graph recovered: 32% (PC), 66% (GES) -> 72% | Improved |
Closest papers by meaning
- CausalTab: Pretraining Across Causal Environments for Tabular Causal Discovery (NeurIPS 2026 · similarity 0.89)
- Score-informed Neural Operator for Enhancing Ordering-based Causal Discovery (NeurIPS 2025 · similarity 0.89)
- Hybrid Top-Down Global Causal Discovery with Local Search for Linear and Nonlinear Additive Noise Models (NeurIPS 2024 · similarity 0.88)
- Differentiable Causal Discovery for Latent Hierarchical Causal Models (ICLR 2025 · similarity 0.88)
- Learning Structural Causal Models from Ordering: Identifiable Flow Models (AAAI 2025 · similarity 0.88)
- On Causal Discovery in the Presence of Deterministic Relations (NeurIPS 2024 · 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.
