A Universal Marginalizer for Amortized Inference in Generative Models

Abstract

We consider the problem of inference in a causal generative model where the set of available observations differs between data instances. We show how combining samples drawn from the graphical model with an appropriate masking function makes it possible to train a single neural network to approximate all the corresponding conditional marginal distributions and thus amortize the cost of inference. We further demonstrate that the efficiency of importance sampling may be improved by basing proposals on the output of the neural network. We also outline how the same network can be used to generate samples from an approximate joint posterior via a chain decomposition of the graph.

Submitted to the NIPS 2017 Workshop on Advances in Approximate Bayesian Inference

References in corpus (2)

• Adam: A Method for Stochastic Optimization
• Recognition Networks for Approximate Inference in BN20 Networks

Cited by in corpus (11)

• Explaining by Removing: A Unified Framework for Model Explanation
• Variational Autoencoder with Arbitrary Conditioning
• Learning Neural Causal Models from Unknown Interventions
• Shapley explainability on the data manifold
• Flow Models for Arbitrary Conditional Likelihoods
• Learning about an exponential amount of conditional distributions
• Adversarially-learned Inference via an Ensemble of Discrete Undirected Graphical Models
• Arbitrary Conditional Distributions with Energy
• Using Shapley Values and Variational Autoencoders to Explain Predictive Models with Dependent Mixed Features
• Inference Over Programs That Make Predictions
• Partially Observed Exchangeable Modeling