AI Paper Atlas

Research map / Probabilistic and scientific modeling / Causal discovery

Global Directional Priors with Local Statistical Validation for Scalable Causal Discovery

Wei Yuan, Zixuan Shao, Shuhui Wang

ICML 2026 · Probabilistic and scientific modeling › Causal discovery · approach: Constraint-based structure learning with latents

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 changedWhereMeasuredEffect
use OCMB-HITON instead of PCPC · Scale-free graphs (d=100)SHD: 158.7 -> 60.3Improved
add OCMB refinement on top of CaPS orderingCaPS-alone · Scale-free graphs (d=100)SHD: 446.0 -> 65.3Improved
replace IAMB with HITON-MB as Stage II learner (ablation)OCMB · Scale-free graphs (d=100-200)F1: 0.489 -> 0.542Improved
replace IAMB with MMPC as Stage II learner (ablation)OCMB · Scale-free graphs (d=100-200)F1: 0.489 -> 0.487No clear effect
use learned CaPS ordering instead of random ordering (ablation)OCMB · Scale-free graphs (d=100)F1: 0.027 -> 0.437Improved
use informative backbone scores instead of random scores (ablation)OCMB · Scale-free graphs (d=100)F1: 0.039 -> 0.437Improved
use oracle ordering instead of CaPS ordering (ablation)OCMB · Scale-free graphs (d=100)F1: 0.437 -> 0.834Improved
corrupt 50% of the oracle ordering (ablation)OCMB · Scale-free graphs (d=100, n=1000)F1: 0.84 -> 0.56Hurt
sweep K/d from 0.02 to 0.40 (ablation)OCMB (CaPS) · ER graphs d=50 and d=200F1: varies by less than 4%No clear effect
add OCMB constraints versus unconstrained PC under query budgetPC · Scale-free graphs (d=100, n=2000)SHD: 112.4 -> 62.8Improved
add OCMB refinement on CaPS ordering at d=500CaPS ordering · Scale-free graphs (d=500, 1/5 seeds)SHD: 12,361 -> 328Improved
use OCMB-SciNO instead of EEMBIEEMBI · DREAM3 Ecoli1SHD: 124 -> 70Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy