Towards a Neural Statistician

TL;DR

This paper introduces a neural network model called a neural statistician that learns to compute dataset-level statistics in an unsupervised manner, enabling efficient transfer learning, clustering, and classification across datasets.

Contribution

It extends variational autoencoders to learn dataset representations, facilitating unsupervised dataset summarization and transfer learning for new datasets.

Findings

Learned statistics enable dataset clustering.

Transferred generative models to new datasets.

Classified unseen classes using dataset summaries.

Abstract

An efficient learner is one who reuses what they already know to tackle a new problem. For a machine learner, this means understanding the similarities amongst datasets. In order to do this, one must take seriously the idea of working with datasets, rather than datapoints, as the key objects to model. Towards this goal, we demonstrate an extension of a variational autoencoder that can learn a method for computing representations, or statistics, of datasets in an unsupervised fashion. The network is trained to produce statistics that encapsulate a generative model for each dataset. Hence the network enables efficient learning from new datasets for both unsupervised and supervised tasks. We show that we are able to learn statistics that can be used for: clustering datasets, transferring generative models to new datasets, selecting representative samples of datasets and classifying previously unseen classes. We refer to our model as a neural statistician, and by this we mean a neural network that can learn to compute summary statistics of datasets without supervision.