Unsupervised vs. transfer learning for multimodal one-shot matching of speech and images

TL;DR

This paper compares unsupervised and transfer learning methods for one-shot speech-image matching, finding transfer learning generally outperforms unsupervised approaches on paired digit datasets.

Contribution

It provides a systematic comparison between unsupervised and transfer learning for multimodal one-shot matching, highlighting the advantages of transfer learning.

Findings

Transfer learning outperforms unsupervised models in one-shot matching.

Unsupervised autoencoder-like models are less effective than supervised classifiers.

Combining unsupervised and transfer learning does not significantly improve performance.

Abstract

We consider the task of multimodal one-shot speech-image matching. An agent is shown a picture along with a spoken word describing the object in the picture, e.g. cookie, broccoli and ice-cream. After observing one paired speech-image example per class, it is shown a new set of unseen pictures, and asked to pick the "ice-cream". Previous work attempted to tackle this problem using transfer learning: supervised models are trained on labelled background data not containing any of the one-shot classes. Here we compare transfer learning to unsupervised models trained on unlabelled in-domain data. On a dataset of paired isolated spoken and visual digits, we specifically compare unsupervised autoencoder-like models to supervised classifier and Siamese neural networks. In both unimodal and multimodal few-shot matching experiments, we find that transfer learning outperforms unsupervised training. We also present experiments towards combining the two methodologies, but find that transfer learning still performs best (despite idealised experiments showing the benefits of unsupervised learning).