Attentive Recurrent Comparators

TL;DR

This paper introduces Attentive Recurrent Comparators (ARCs), a novel model for rapid learning that forms object representations through observation cycles, achieving state-of-the-art one-shot classification on Omniglot with super-human accuracy.

Contribution

The paper presents ARCs, a new model that cycles through objects to form representations, enabling effective one-shot learning from raw pixel data.

Findings

Achieved 1.5% error rate on Omniglot one-shot classification.

First super-human result for this task using a generic pixel-based model.

Demonstrated the effectiveness of ARCs in forming dynamic representations.

Abstract

Rapid learning requires flexible representations to quickly adopt to new evidence. We develop a novel class of models called Attentive Recurrent Comparators (ARCs) that form representations of objects by cycling through them and making observations. Using the representations extracted by ARCs, we develop a way of approximating a \textit{dynamic representation space} and use it for one-shot learning. In the task of one-shot classification on the Omniglot dataset, we achieve the state of the art performance with an error rate of 1.5\%. This represents the first super-human result achieved for this task with a generic model that uses only pixel information.