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.
%0 Generic
%1 liu2018darts
%A Liu, Hanxiao
%A Simonyan, Karen
%A Yang, Yiming
%D 2018
%K 2018 architecture arxiv deep-learning neural-networks search
%T DARTS: Differentiable Architecture Search
%U http://arxiv.org/abs/1806.09055
%X 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.
@misc{liu2018darts,
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.},
added-at = {2018-06-27T08:21:58.000+0200},
author = {Liu, Hanxiao and Simonyan, Karen and Yang, Yiming},
biburl = {https://www.bibsonomy.org/bibtex/229bb5401cf3218cac8120d51fdf465d3/achakraborty},
description = {[1806.09055] DARTS: Differentiable Architecture Search},
interhash = {29af81f097d613a69fdffa7564cd4870},
intrahash = {29bb5401cf3218cac8120d51fdf465d3},
keywords = {2018 architecture arxiv deep-learning neural-networks search},
note = {cite arxiv:1806.09055},
timestamp = {2018-06-27T08:22:16.000+0200},
title = {DARTS: Differentiable Architecture Search},
url = {http://arxiv.org/abs/1806.09055},
year = 2018
}