Learning Latent Structural Causal Models
This addresses the challenge of causal learning in machine learning tasks where only low-level data is available, enabling better explanations and generalization, though it is incremental as it builds on existing SCM frameworks with a new inference approach.
The paper tackles the problem of learning latent structural causal models (SCMs) from low-level data like images, where high-level causal variables are unobserved, by proposing a Bayesian inference method for linear Gaussian additive noise SCMs. The result demonstrates efficacy on synthetic and causally generated image datasets, including image generation from unseen interventions to verify out-of-distribution generalization.
Causal learning has long concerned itself with the accurate recovery of underlying causal mechanisms. Such causal modelling enables better explanations of out-of-distribution data. Prior works on causal learning assume that the high-level causal variables are given. However, in machine learning tasks, one often operates on low-level data like image pixels or high-dimensional vectors. In such settings, the entire Structural Causal Model (SCM) -- structure, parameters, \textit{and} high-level causal variables -- is unobserved and needs to be learnt from low-level data. We treat this problem as Bayesian inference of the latent SCM, given low-level data. For linear Gaussian additive noise SCMs, we present a tractable approximate inference method which performs joint inference over the causal variables, structure and parameters of the latent SCM from random, known interventions. Experiments are performed on synthetic datasets and a causally generated image dataset to demonstrate the efficacy of our approach. We also perform image generation from unseen interventions, thereby verifying out of distribution generalization for the proposed causal model.