Research map / Probabilistic and scientific modeling / Causal discovery
BayesDAG: Gradient-Based Posterior Inference for Causal Discovery
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 BayesDAG instead of BCD | Nonlinear causal discovery · ER nonlinear d=70 | E-SHD: 217.05 ± 9.58 -> 143.70 ± 11.61 | Helped |
| Use BayesDAG instead of BCD | Nonlinear causal discovery · ER nonlinear d=100 | E-SHD: 362.66 ± 29.18 -> 295.92 ± 24.67 | Helped |
| Use BayesDAG instead of BCD | Causal discovery on semi-synthetic data · SynTReN d=20 | E-SHD: 38.38 ± 7.12 -> 34.21 ± 2.82 | Helped |
| Use BayesDAG instead of BGES | Causal discovery on semi-synthetic data · SynTReN d=20 | Edge F1: 0.21 ± 0.05 -> 0.20 ± 0.02 | Hurt |
| Use BayesDAG instead of BGES | Causal discovery on real-world data · Sachs Protein Cells d=11 | E-SHD: 16.61 ± 0.44 -> 18.92 ± 1.0 | Hurt |
| Use BayesDAG instead of BGES | Causal discovery on real-world data · Sachs Protein Cells d=11 | Edge F1: 0.22 ± 0.02 -> 0.26 ± 0.04 | Helped |
| Use a larger initialized node-potential scale (ablation) | BayesDAG · Nonlinear ER d=30 | Posterior inference performance | Hurt |
| Increase the number of parallel SG-MCMC chains (ablation) | BayesDAG · Nonlinear ER d=30 | E-SHD | No clear effect |
| Increase injected SG-MCMC noise for node potentials (ablation) | BayesDAG · Nonlinear ER d=30 | Posterior inference performance | No clear effect |
| Inject SG-MCMC noise rather than using no injected noise (ablation) | BayesDAG · Nonlinear ER d=30 | Posterior inference performance | Helped |
| Increase injected SG-MCMC noise for model parameters (ablation) | BayesDAG · Nonlinear ER d=30 | Posterior inference performance | Hurt |
| Replace the SG-MCMC+VI graph inference with SG-MCMC over a continuous relaxation of W (ablation) | BayesDAG · Reported experiments | Empirical performance | Hurt |
Closest papers by meaning
- A Meta-Learning Approach to Bayesian Causal Discovery (ICLR 2025 · similarity 0.92)
- Sample Efficient Bayesian Learning of Causal Graphs from Interventions (NeurIPS 2024 · similarity 0.92)
- Causal Discovery via Bayesian Optimization (ICLR 2025 · similarity 0.90)
- CausalTab: Pretraining Across Causal Environments for Tabular Causal Discovery (NeurIPS 2026 · similarity 0.90)
- Bivariate Causal Discovery using Bayesian Model Selection (ICML 2024 · similarity 0.90)
- Challenges and Considerations in the Evaluation of Bayesian Causal Discovery (ICML 2024 · similarity 0.90)
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.