Practical do-Shapley Explanations with Estimand-Agnostic Causal Inference

2025/09/24 by Álvaro Parafita, Tomas Garriga, Parafita, Álvaro +5 · 1 voice
Computer Science · #FOS: Computer and information sciences #Machine Learning (cs.LG) #cs.LG

paper · pdf · doi:10.48550/arxiv.2509.20211

arxiv published 2025/09/24 · arxiv updated 2026/01/12

Abstract

Among explainability techniques, SHAP stands out as one of the most popular, but often overlooks the causal structure of the problem. In response, do-SHAP employs interventional queries, but its reliance on estimands hinders its practical application. To address this problem, we propose the use of estimand-agnostic approaches, which allow for the estimation of any identifiable query from a single model, making do-SHAP feasible on complex graphs. We also develop a novel algorithm to significantly accelerate its computation at a negligible cost, as well as a method to explain inaccessible Data Generating Processes. We demonstrate the estimation and computational performance of our approach, and validate it on two real-world datasets, highlighting its potential in obtaining reliable explanations.

Citations

• Causal normalizing flows: from theory to practice
• Efficient SAGE Estimation via Causal Structure Learning
• Spurious Features Everywhere -- Large-Scale Detection of Harmful Spurious Features in ImageNet
• Algorithms to estimate Shapley value feature attributions
• Generative Adversarial Networks
• Algorithmic Causal Effect Identification with causaleffect
• The Causal-Neural Connection: Expressiveness, Learnability, and Inference
• Toward Causal Representation Learning
• Causal Shapley Values: Exploiting Causal Knowledge to Explain Individual Predictions of Complex Models
• Shapley Flow: A Graph-based Approach to Interpreting Model Predictions
• Denoising Diffusion Probabilistic Models
• Deep Structural Causal Models for Tractable Counterfactual Inference
• Normalizing Flows for Probabilistic Modeling and Inference
• Feature relevance quantification in explainable AI: A causal problem
• Asymmetric Shapley values: incorporating causal knowledge into model-agnostic explainability
• Explaining Visual Models by Causal Attribution
• Neural Spline Flows
• Graph Neural Networks: A Review of Methods and Applications
• Graph neural networks: A review of methods and applications
• CausalGAN: Learning Causal Implicit Generative Models with Adversarial Training
• A Unified Approach to Interpreting Model Predictions
• Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)
• Intriguing properties of neural networks
• Identification of Conditional Interventional Distributions
• Modeling Causal Mechanisms with Diffusion Models for Interventional and Counterfactual Queries

Discussions

• Thrilled to announce that our paper “Practical do-Shapley Explanations with Estimand-Agnostic Causal Inference” has been accepted at #NeurIPS2025 with spotlight! 🎉 🔗 arxiv.org/abs/2509.20211 If you' [bsky, 4 points, 1 comments]

Related

• Computing SHAP Efficiently Using Model Structure Information
• Generalized SHAP: Generating multiple types of explanations in machine learning
• Causal Shapley Values: Exploiting Causal Knowledge to Explain Individual Predictions of Complex Models
• On the Computational Tractability of the (Many) Shapley Values
• From Shapley Values to Generalized Additive Models and back