The Variational Homoencoder: Learning to learn high capacity generative models from few examples

TL;DR

The paper introduces the Variational Homoencoder (VHE), a hierarchical latent variable model that improves learning in high-capacity neural networks like PixelCNN, excelling in few-shot learning, generation, and classification tasks.

Contribution

It proposes the VHE framework, a modified variational autoencoder that better utilizes latent variables for hierarchical generative modeling with neural networks.

Findings

VHE outperforms existing models on Omniglot likelihood.

Achieves strong one-shot classification and generation performance.

Validates on natural images from YouTube Faces database.

Abstract

Hierarchical Bayesian methods can unify many related tasks (e.g. k-shot classification, conditional and unconditional generation) as inference within a single generative model. However, when this generative model is expressed as a powerful neural network such as a PixelCNN, we show that existing learning techniques typically fail to effectively use latent variables. To address this, we develop a modification of the Variational Autoencoder in which encoded observations are decoded to new elements from the same class. This technique, which we call a Variational Homoencoder (VHE), produces a hierarchical latent variable model which better utilises latent variables. We use the VHE framework to learn a hierarchical PixelCNN on the Omniglot dataset, which outperforms all existing models on test set likelihood and achieves strong performance on one-shot generation and classification tasks. We additionally validate the VHE on natural images from the YouTube Faces database. Finally, we develop extensions of the model that apply to richer dataset structures such as factorial and hierarchical categories.