CARS: Continuous Evolution for Efficient Neural Architecture Search

TL;DR

This paper introduces a continuous evolutionary neural architecture search method that efficiently generates high-performing networks with diverse sizes in just 0.4 GPU days, outperforming state-of-the-art models on ImageNet.

Contribution

It proposes a novel continuous evolutionary approach for NAS that accelerates search and produces multiple optimized architectures efficiently.

Findings

Networks with 3.7M to 5.1M parameters outperform state-of-the-art on ImageNet.

The method achieves high efficiency with only 0.4 GPU days.

Multiple architectures with different sizes and performances are generated.

Abstract

Searching techniques in most of existing neural architecture search (NAS) algorithms are mainly dominated by differentiable methods for the efficiency reason. In contrast, we develop an efficient continuous evolutionary approach for searching neural networks. Architectures in the population that share parameters within one SuperNet in the latest generation will be tuned over the training dataset with a few epochs. The searching in the next evolution generation will directly inherit both the SuperNet and the population, which accelerates the optimal network generation. The non-dominated sorting strategy is further applied to preserve only results on the Pareto front for accurately updating the SuperNet. Several neural networks with different model sizes and performances will be produced after the continuous search with only 0.4 GPU days. As a result, our framework provides a series of networks with the number of parameters ranging from 3.7M to 5.1M under mobile settings. These networks surpass those produced by the state-of-the-art methods on the benchmark ImageNet dataset.