AI Paper Atlas

Research map / Probabilistic and scientific modeling / Causal discovery

Shapley Regularized Neural Granger Causality

Maolin Yang, Zhoufan Zhu, Yuanhe Tian, Kun Gao, MUYI LI

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
replace local Jacobian importance with global Info-Shap in the regularizer (l1 form) (ablation)Residual MLP neural Granger causality · Lorenz-96 (d=50, F=40, T=1000)AUROC: 0.941 -> 0.984Improved
replace Jacobian with Info-Shap on linear VAR(3) (ablation)Residual MLP neural Granger causality · VAR(3) (d=50, T=500)AUROC: 0.895 -> 0.898No clear effect
use Frobenius-norm random-projection variants instead of exact l1 penalties (ablation)Info-Shap and Jacobian regularizers · Lorenz-96 (d=50, F=40, T=1000)AUROC: 0.984 -> 0.775 (Shap -> F-Shap)Hurt
use exact Shap instead of F-Shap on small-sample DREAM3 (ablation)SRNGC · DREAM3 Ecoli-1AUROC: 0.632 -> 0.700Improved
use F-Shap instead of exact Shap on larger-sample CausalTime (ablation)SRNGC · CausalTime AQIAUROC: 0.946 -> 0.961Improved
use joint modeling instead of component-wise modeling (ablation)Shap regularizer with MLP backbone · fMRI first 5 subjectsAUROC: 0.854 -> 0.874Improved
use joint modeling instead of component-wise modeling (ablation)Shap regularizer with LSTM backbone · fMRI first 5 subjectsAUROC: 0.705 -> 0.847Improved
use model-agnostic Shap regularizer instead of layer-weight importance in component-wise cMLP (ablation)cMLP · fMRI first 5 subjectsAUROC: 0.726 -> 0.854Improved
use F-Shap instead of Shap with LSTM joint model (ablation)LSTM · fMRI first 5 subjectsAUROC: 0.847 -> 0.713Hurt
apply Info-Shap on the Medical dataset with the same Residual MLPSRNGC / F-SRNGC · CausalTime MedicalAUROC: SRNGC 0.720 vs CUTS+ 0.820Hurt
use knockoff-style thresholding with q = 0.01 and an over-specified lag order of 4SRNGC · VAR(3) (d=10, T=1000)graph recoveryImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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