A Relational Model for One-Shot Classification

TL;DR

This paper introduces a deep learning model with relational inductive bias that achieves state-of-the-art one-shot classification performance, surpassing humans and previous methods without data augmentation.

Contribution

It presents a novel relational matching model for one-shot classification that leverages local and pairwise attention, improving sample efficiency.

Findings

Achieves perfect accuracy on Omniglot one-shot task

Exceeds human-level performance in one-shot classification

Outperforms previous state-of-the-art methods without data augmentation

Abstract

We show that a deep learning model with built-in relational inductive bias can bring benefits to sample-efficient learning, without relying on extensive data augmentation. The proposed one-shot classification model performs relational matching of a pair of inputs in the form of local and pairwise attention. Our approach solves perfectly the one-shot image classification Omniglot challenge. Our model exceeds human level accuracy, as well as the previous state of the art, with no data augmentation.