Neural Predictor for Neural Architecture Search

TL;DR

This paper introduces a simple, three-step neural predictor approach for neural architecture search that is highly sample-efficient and competitive with complex methods on benchmarks like NASBench-101 and ImageNet.

Contribution

The paper presents a straightforward regression-based neural predictor method that significantly improves sample efficiency in neural architecture search.

Findings

Over 20 times more sample efficient than Regularized Evolution on NASBench-101

Achieves competitive results on ImageNet with simpler approach

Outperforms complex weight sharing methods in efficiency

Abstract

Neural Architecture Search methods are effective but often use complex algorithms to come up with the best architecture. We propose an approach with three basic steps that is conceptually much simpler. First we train N random architectures to generate N (architecture, validation accuracy) pairs and use them to train a regression model that predicts accuracy based on the architecture. Next, we use this regression model to predict the validation accuracies of a large number of random architectures. Finally, we train the top-K predicted architectures and deploy the model with the best validation result. While this approach seems simple, it is more than 20 times as sample efficient as Regularized Evolution on the NASBench-101 benchmark and can compete on ImageNet with more complex approaches based on weight sharing, such as ProxylessNAS.