A Language for Counterfactual Generative Models

Abstract

Probabilistic programming languages provide syntax to define and condition generative models but lack mechanisms for counterfactual queries. We introduce OmegaC : a causal probabilistic programming language for constructing and performing inference in counterfactual generative models. In OmegaC , a counterfactual generative model is a program that combines both conditioning and causal interventions to represent queries such as “given that X is true, what if Y were the case?“. We define the syntax and semantics of OmegaC and demonstrate examples in population dynamics, inverse planning and causation.

Publication
In submission
Date