DARTS: Differentiable Architecture Search

TL;DR

DARTS introduces a differentiable approach to neural architecture search, enabling efficient gradient-based optimization for designing high-performance models in image classification and language modeling tasks.

Contribution

It presents a novel continuous relaxation of architecture search space, allowing gradient descent to efficiently discover neural architectures, significantly faster than previous methods.

Findings

Achieves state-of-the-art results on CIFAR-10 and ImageNet.

Demonstrates high-quality architectures for language modeling.

Orders of magnitude faster than non-differentiable techniques.

Abstract

This paper addresses the scalability challenge of architecture search by formulating the task in a differentiable manner. Unlike conventional approaches of applying evolution or reinforcement learning over a discrete and non-differentiable search space, our method is based on the continuous relaxation of the architecture representation, allowing efficient search of the architecture using gradient descent. Extensive experiments on CIFAR-10, ImageNet, Penn Treebank and WikiText-2 show that our algorithm excels in discovering high-performance convolutional architectures for image classification and recurrent architectures for language modeling, while being orders of magnitude faster than state-of-the-art non-differentiable techniques. Our implementation has been made publicly available to facilitate further research on efficient architecture search algorithms.