Counterfactual Sum-Product Networks

Florian Peter Busch, Moritz Willig, Matej Zečević, Kristian Kersting, Devendra Singh Dhami

Keywords: sum-product networks, counterfactuals, causality

Abstract:

The complexity and vastness of our world can require large models with numerous variables. Unfortunately, coming up with a model that is both accurate and able to provide predictions in a reasonable amount of time can prove difficult. One possibility to help overcome such problems is sum-product networks (SPNs), probabilistic models with the ability to tractably perform inference in linear time.
In this paper, we extend SPNs' capabilities to the field of causality and introduce counterfactual Sum-Product Networks (cf-SPNs), a type of SPNs capable of answering counterfactual questions. cf-SPNs make use of a neural component that sets the parameters of an SPN such that it represents the specified counterfactual world. We show that cf-SPNs can successfully learn counterfactual distributions.

Submission Number: 16