Interventional Sum-Product Networks: Causal Inference with Tractable\n Probabilistic Models

2021/02/20 by Matej Zečević, Devendra Singh Dhami, Zečević, Matej +7 · 2 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · #Bayesian Modeling and Causal Inference #Biomedical Text Mining and Ontologies #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning and Algorithms #Machine Learning in Healthcare

paper · pdf · doi:10.48550/arxiv.2102.10440

openalex publication_date 2021/02/20 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28

Abstract

While probabilistic models are an important tool for studying causality,\ndoing so suffers from the intractability of inference. As a step towards\ntractable causal models, we consider the problem of learning interventional\ndistributions using sum-product networks (SPNs) that are over-parameterized by\ngate functions, e.g., neural networks. Providing an arbitrarily intervened\ncausal graph as input, effectively subsuming Pearl's do-operator, the gate\nfunction predicts the parameters of the SPN. The resulting interventional SPNs\nare motivated and illustrated by a structural causal model themed around\npersonal health. Our empirical evaluation on three benchmark data sets as well\nas a synthetic health data set clearly demonstrates that interventional SPNs\nindeed are both expressive in modelling and flexible in adapting to the\ninterventions.\n

Citations

• Investigating Causal Relations by Econometric Models and Cross-spectral Methods
• Causal Reasoning from Meta-reinforcement Learning
• Conditional Sum-Product Networks: Imposing Structure on Deep\n Probabilistic Architectures
• Learning Neural Causal Models from Unknown Interventions
• Causality for Machine Learning
• CausalML: Python Package for Causal Machine Learning
• Sum-product networks: A survey
• DoWhy: An End-to-End Library for Causal Inference

Cited by

• Tagged for Direction: Pinning Down Causal Edge Directions with Precision
• Scaling Probabilistic Circuits via Monarch Matrices

Related

• χSPN: Characteristic Interventional Sum-Product Networks for Causal Inference in Hybrid Domains
• Interventions and Counterfactuals in Tractable Probabilistic Models: Limitations of Contemporary Transformations
• Treatment-RSPN: Recurrent Sum-Product Networks for Sequential Treatment Regimes
• Estimating Interventional Distributions with Uncertain Causal Graphs through Meta-Learning
• Sum-product networks: A survey