Moment-Matching Graph-Networks for Causal Inference
This work addresses causal inference for researchers and practitioners by enabling automated sampling of latent space conditional probability distributions for graphical interventions, though it appears incremental as it builds on existing autoencoder methods.
The paper tackles the problem of simulating non-linear structural equation models from observational data by proposing a fully unsupervised deep-learning framework that uses moment-matching loss functions on causal Bayesian graphs, resulting in generative conditional-moment-matching graph-neural-networks capable of generating out-of-sample interventional probabilities faithful to ground truth distributions beyond the training set range.
In this note we explore a fully unsupervised deep-learning framework for simulating non-linear structural equation models from observational training data. The main contribution of this note is an architecture for applying moment-matching loss functions to the edges of a causal Bayesian graph, resulting in a generative conditional-moment-matching graph-neural-network. This framework thus enables automated sampling of latent space conditional probability distributions for various graphical interventions, and is capable of generating out-of-sample interventional probabilities that are often faithful to the ground truth distributions well beyond the range contained in the training set. These methods could in principle be used in conjunction with any existing autoencoder that produces a latent space representation containing causal graph structures.