Interventional Sum-Product Networks: Causal Inference with Tractable Probabilistic Models
This work addresses the challenge of tractable causal inference for researchers in probabilistic modeling and causality, representing an incremental advancement by adapting existing SPN methods to incorporate interventions.
The paper tackles the problem of intractable inference in causal models by introducing interventional sum-product networks (SPNs) that use neural network gate functions to predict SPN parameters for arbitrary interventions, demonstrating their expressiveness and flexibility on three benchmark datasets and a synthetic health dataset.
While probabilistic models are an important tool for studying causality, doing so suffers from the intractability of inference. As a step towards tractable causal models, we consider the problem of learning interventional distributions using sum-product networks (SPNs) that are over-parameterized by gate functions, e.g., neural networks. Providing an arbitrarily intervened causal graph as input, effectively subsuming Pearl's do-operator, the gate function predicts the parameters of the SPN. The resulting interventional SPNs are motivated and illustrated by a structural causal model themed around personal health. Our empirical evaluation on three benchmark data sets as well as a synthetic health data set clearly demonstrates that interventional SPNs indeed are both expressive in modelling and flexible in adapting to the interventions.