Variational Autoencoder with Arbitrary Conditioning

TL;DR

This paper introduces a variational autoencoder model capable of conditioning on any subset of features to generate the remaining features in a single step, applicable to real-valued and categorical data.

Contribution

It presents a novel neural probabilistic model that allows arbitrary feature conditioning and efficient sampling, trained via stochastic variational Bayes.

Findings

Effective on synthetic data, feature imputation, and image inpainting

Generates diverse samples

Outperforms existing methods in conditional generation

Abstract

We propose a single neural probabilistic model based on variational autoencoder that can be conditioned on an arbitrary subset of observed features and then sample the remaining features in "one shot". The features may be both real-valued and categorical. Training of the model is performed by stochastic variational Bayes. The experimental evaluation on synthetic data, as well as feature imputation and image inpainting problems, shows the effectiveness of the proposed approach and diversity of the generated samples.