Research map / Probabilistic and scientific modeling / Causal discovery
Trust Your $\nabla$: Gradient-based Intervention Targeting for Causal Discovery
Mateusz Olko, Michał Zając, Aleksandra Nowak, Nino Scherrer, Yashas Annadani, Stefan Bauer, Łukasz Kuciński, Piotr Miłoś
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 GIT targeting instead of Random | ENCO · 24 setups, 12 graphs x 2 regimes | mean AUSHD best or comparable count: 12 -> 18 | Improved |
| use GIT targeting instead of Random | ENCO · 24 setups | mean SHD best or comparable count: 22 -> 17 | Mixed |
| use GIT instead of AIT | ENCO · 24 setups | mean AUSHD best or comparable count: 6 -> 18 | Improved |
| use GIT instead of CBED | ENCO · 24 setups | mean AUSHD best or comparable count: 6 -> 18 | Improved |
| restrict to low data regime N=1056 | GIT · 12 graphs | better or comparable to other non-privileged methods: 11 out of 12 graphs | Improved |
| replace model-imagined data with real interventional data (GIT-privileged) (ablation) | GIT · 24 setups | mean AUSHD best or comparable count: 18 -> 24 | Improved |
| use model-sampled data instead of real data in GIT score, low vs regular regime (ablation) | GIT · 24 setups | equivalence with GIT-privileged: 10 cases low data, 5 cases regular | Mixed |
| increase interventional batch size from 32 to 1024 | GIT · synthetic 25-node graphs | AUSHD vs baselines and final SHD: SHD 0 on all graphs; collider 5.0 vs Random 9.9 | Improved |
| increase interventional batch size from 32 to 1024 | CBED and AIT · synthetic 25-node graphs | AUSHD vs Random: collider 9.9 -> 11.4 | Hurt |
| increase Monte-Carlo sample size for GIT score (ablation) | GIT · ENCO | performance | No clear effect |
| use GIT on graphs with singleton Markov equivalence class (cancer, earthquake) | GIT · real-world graphs | AUSHD | No clear effect |
| GIT vs GIT-privileged intervention choices | GIT · real-world graphs | Spearman correlation of target choices: 73% | Improved |
Closest papers by meaning
- CausalTab: Pretraining Across Causal Environments for Tabular Causal Discovery (NeurIPS 2026 · similarity 0.88)
- Identifying biological perturbation targets through causal differential networks (ICML 2025 · similarity 0.87)
- Bayesian Active Learning for Bivariate Causal Discovery (ICML 2025 · similarity 0.87)
- Sample Efficient Bayesian Learning of Causal Graphs from Interventions (NeurIPS 2024 · similarity 0.86)
- Causal Discovery by Interventions via Integer Programming (AAAI 2025 · similarity 0.86)
- On Causal Discovery in the Presence of Deterministic Relations (NeurIPS 2024 · similarity 0.86)
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.
