Estimation of a Causal Directed Acyclic Graph Process using Non-Gaussianity

TL;DR

This paper introduces CGP-LiNGAM, a new causal modeling approach that reduces parameters and simplifies causal graph inference in time series data by leveraging Graph Signal Processing techniques.

Contribution

The paper proposes CGP-LiNGAM, a novel method that simplifies causal inference in time series by reducing model complexity and using GSP for a unified causal graph.

Findings

CGP-LiNGAM has fewer parameters than VAR-LiNGAM.

It effectively infers causal relations with a single graph.

The approach improves interpretability and efficiency.

Abstract

Numerous approaches have been proposed to discover causal dependencies in machine learning and data mining; among them, the state-of-the-art VAR-LiNGAM (short for Vector Auto-Regressive Linear Non-Gaussian Acyclic Model) is a desirable approach to reveal both the instantaneous and time-lagged relationships. However, all the obtained VAR matrices need to be analyzed to infer the final causal graph, leading to a rise in the number of parameters. To address this issue, we propose the CGP-LiNGAM (short for Causal Graph Process-LiNGAM), which has significantly fewer model parameters and deals with only one causal graph for interpreting the causal relations by exploiting Graph Signal Processing (GSP).