Deep Clustering for Domain Adaptation

We address the heterogeneous domain adaptation task: adapting a classifier trained on data from one domain to operate on another domain that also has a different label space. We consider two settings that both exhibit label scarcity of some form—one where only unlabelled data is available, and another where a small volume of labelled data is available in addition to the unlabelled data. Our method is based on two specialisations of a recently proposed approach for deep clustering. It is shown that our approach noticeably outperforms other methods based on deep clustering in both the fully unsupervised and the semi-supervised settings.

1. INTRODUCTION

Domain adaptation aims to alleviate the need for labelled examples in a given target domain using knowledge transferred from a related source domain, such as across image data of different camera types [1]. Domain adaptation (DA) is a very well studied topic with numerous competing methods [1]. Clustering-based DA approaches group unlabelled target domain data examples that are likely to belong to the same class. One area of DA where clustering methods are particularly useful is that of heterogeneous domain adaptation—where the target domain contains novel classes compared to the source domain. For DA problems of this type, clustering-based DA methods [2], [3] provide one of the only solutions.