Research map / Probabilistic and scientific modeling / Causal discovery
Gradient-Based Neural DAG 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 nonlinear neural conditionals with a continuous acyclicity constraint | GraN-DAG versus DAG-GNN · Gauss-ANM, ER1, 10 nodes | SHD: 11.4$\pm$3.1 -> 1.7$\pm$2.5 | Helped |
| Use nonlinear neural conditionals with a continuous acyclicity constraint | GraN-DAG versus DAG-GNN · Gauss-ANM, ER1, 10 nodes | SID: 37.6$\pm$14.4 -> 1.7$\pm$3.1 | Helped |
| Add PNS and pruning (ablation) | GraN-DAG · Gauss-ANM, ER1, 50 nodes | SHD: 1086.8$\pm$48.8 -> 6.1$\pm$3.3 | Helped |
| Add PNS and pruning (ablation) | GraN-DAG · Gauss-ANM, ER1, 50 nodes | SID: 31.6$\pm$23.6 -> 29.3$\pm$19.5 | Mixed |
| Add PNS without pruning (ablation) | GraN-DAG · Gauss-ANM, ER1, 50 nodes | SHD: 1086.8$\pm$48.8 -> 540.4$\pm$70.3 | Helped |
| Add pruning without PNS (ablation) | GraN-DAG · Gauss-ANM, ER1, 50 nodes | SHD: 1086.8$\pm$48.8 -> 11.8$\pm$5.0 | Helped |
| Allow conditional variance to depend on parent variables (ablation) | GraN-DAG (GraN-DAG++) · Protein signaling data set | SHD: 13 -> 13 | No clear effect |
| Allow conditional variance to depend on parent variables (ablation) | GraN-DAG (GraN-DAG++) · SynTReN, 20 nodes | SID: 161.7$\pm$53.4 -> 127.5$\pm$52.8 | Mixed |
| Use the expected Jacobian rather than weighted adjacency strengths for final edge removal (ablation) | GraN-DAG · Structure learning experiments | graph recovery | Helped |
| Increase sample size from 500 to 5000 (ablation) | GraN-DAG versus CAM · Gauss-ANM, ER4, 50 nodes | SID: CAM 845.0 $\pm$ 159.8; GraN-DAG 554.1 $\pm$ 117.9 | Helped |
| Apply PNS and pruning to NOTEARS and DAG-GNN (ablation) | NOTEARS and DAG-GNN · Gauss-ANM, ER1, 50 nodes | SID: DAG-GNN 322.9$\pm$103.8 -> 316.5$\pm$106.9; NOTEARS 327.1$\pm$110.9 -> 326.4$\pm$106.9 | No clear effect |
Closest papers by meaning
- Causal Discovery via Bayesian Optimization (ICLR 2025 · similarity 0.90)
- Analytic DAG Constraints for Differentiable DAG Learning (ICLR 2025 · similarity 0.90)
- CoLiDE: Concomitant Linear DAG Estimation (ICLR 2024 · similarity 0.89)
- Effective Causal Discovery under Identifiable Heteroscedastic Noise Model (AAAI 2024 · similarity 0.89)
- Differentiable Structure Learning with Ancestral Constraints (ICML 2025 · similarity 0.87)
- Boosting Causal Structure Learning: An Asymmetric Exponential Modulation Gaussian-Based Adaptive Sample Reweighting Framework (AAAI 2025 · 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.