AI Paper Atlas

Research map / Probabilistic and scientific modeling / Causal discovery

Use What You Know: Causal Foundation Models with Partial Graphs

Arik Reuter, Anish Dhir, Cristiana Diaconu, Jake Robertson, Ole Ossen, Frank Hutter, Adrian Weller, Mark van der Wilk, Bernhard Schölkopf

ICML 2026 · Probabilistic and scientific modeling › Causal discovery · approach: Interventional causal discovery

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
condition on partially known ancestral matrix encoding unconfoundedness (ablation)CFM on complex prior · RealCause IHDPsqrt(PEHE): 6.28 -> 5.49Improved
condition on partially known ancestral matrix (ablation)CFM on complex prior · RealCause ACICATE relative error: 0.46 -> 0.17Improved
condition on partially known ancestral matrix (ablation)CFM on complex prior · RealCause CPSsqrt(PEHE): 12800 -> 11213Improved
condition on ancestral information (ablation)CFM on complex prior · RealCause PSID (unbalanced)sqrt(PEHE): 13096 -> 12975Improved
add ancestral information (ATE error) (ablation)CFM on complex prior · RealCause PSID (unbalanced)ATE relative error: 0.98 -> 1.09Hurt
soft learnable attention bias instead of hard masking (ablation)graph-conditioned CFM · linear-Gaussian priorNLL/MSE/R2Improved
add GCN encoder to soft attention bias (ablation)graph-conditioned CFM · complex TFM priorNLLImproved
use ancestor matrix instead of adjacency matrix (ablation)soft-attention conditioning · linear-Gaussian priorMSE and R2: smallImproved
hide most ancestor entries at test time (ablation)graph-conditioned CFM · complex TFM priorMSE and R2Hurt
amortise over varying amounts of ancestral information (ablation)amortised CFM · complex TFM priorperformanceNo clear effect
misspecify edges instead of marking unknown (ablation)graph-conditioned CFM · misspecification studyperformanceHurt

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