A Continuous Structural Intervention Distance to Compare Causal Graphs

Dhanakshirur, M and Laumann, F and Park, J and Barahona, M (2025) A Continuous Structural Intervention Distance to Compare Causal Graphs. [Preprint]

PDF Com_CoM_Inf_Sci_Vol_2200.pdf - Published Version Restricted to Registered users only Download (557kB) | Request a copy

Abstract

Causal inference under interventions requires accurate assessment of differences between true and learned causal graphs. We introduce a new continuous metric that extends beyond graph-based measures like Structural Hamming Distance and Structural Intervention Distance by incorporating underlying data alongside graph structures. Our approach embeds intervention distributions for each node pair as conditional mean embeddings in reproducing kernel Hilbert spaces, then quantifies their disparity using maximum (conditional) mean discrepancy. We present theoretical findings supported by synthetic data experiments. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

Item Type:	Preprint
Publication:	Communications in Computer and Information Science
Publisher:	Springer Science and Business Media Deutschland GmbH
Additional Information:	The copyright for this article belongs to Publishers.
Keywords:	Directed graphs; Hamming distance; Hilbert spaces, Acyclic graphs; Causal graph; Causal inferences; Conditional mean embedding; Conditional means; Directed acyclic graph; Embeddings; Graph distances; Kernel-methods; Structural interventions, Graph embeddings
Department/Centre:	Division of Physical & Mathematical Sciences > Mathematics
Date Deposited:	04 Dec 2024 18:25
Last Modified:	04 Dec 2024 18:25
URI:	http://eprints.iisc.ac.in/id/eprint/86880

Actions (login required)

View Item