AI Paper Atlas

Research map / Probabilistic and scientific modeling / Causal discovery

Causal Score Conditioning for Multi-Resolution Latent Systems

Xuechun Li, Shan Gao, Susu Xu

ICLR 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
add observation-consistency loss term (ablation)SVGDM · Synthetic 3-node (VFO)NRMSE: 0.117 -> 0.103Improved
add causal-blanket score loss term (ablation)SVGDM · Synthetic 3-node (VFO)NRMSE: 0.129 -> 0.103Improved
add local DSM score-matching term (ablation)SVGDM · Synthetic 3-node (VFO)NRMSE: 0.142 -> 0.103Improved
add causal and observation terms to DSM only (ablation)SVGDM · Synthetic 3-node (VFO)NRMSE: 0.154 -> 0.103Improved
use correct causal DAG instead of reversing 1 edge (ablation)SVGDM · syntheticNRMSE: 0.118 -> 0.098Improved
use correct DAG instead of two wrong edges (ablation)SVGDM · syntheticNRMSE: 0.162 -> 0.098Improved
spurious extra edge added to DAG (ablation)SVGDM · syntheticNRMSE: 0.098 -> 0.104Hurt
replace Gaussian noise with Laplace noise (ablation)SVGDM · VFO N=10 sparseNRMSE: 0.1158 -> 0.1167No clear effect
replace Gaussian noise with skewed log-normal mix (ablation)SVGDM · VFO N=10 sparseNRMSE: 0.1158 -> 0.1199Hurt
observe variables at low resolution (LFO) instead of varying resolution (VFO) (ablation)SVGDM · synthetic z1MAPE: 0.0526 -> 0.0756Hurt
scale latent system from 10 to 15 variables (sparse VFO) (ablation)SVGDM · synthetic scalabilityNRMSE: 0.11 -> 0.13Mixed
use causal score diffusion SVGDM instead of multi-view VAE MoPoE-VAEMoPoE-VAE · synthetic z1NRMSE: 0.2241 -> 0.0683Improved

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