Multi-objective Neural Architecture Search with Almost No Training

TL;DR

This paper introduces Random-Weight Evaluation (RWE), a fast method to estimate neural network performance by training only the last layer, enabling efficient multi-objective neural architecture search with minimal computation.

Contribution

The paper presents RWE, a novel approach that drastically reduces NAS evaluation time, allowing for near real-time architecture search without extensive training.

Findings

RWE achieves state-of-the-art architectures on CIFAR-10 in under two hours.

RWE's performance correlates well with fully trained models in ablation studies.

Transfer learning experiments validate RWE's effectiveness on ImageNet.

Abstract

In the recent past, neural architecture search (NAS) has attracted increasing attention from both academia and industries. Despite the steady stream of impressive empirical results, most existing NAS algorithms are computationally prohibitive to execute due to the costly iterations of stochastic gradient descent (SGD) training. In this work, we propose an effective alternative, dubbed Random-Weight Evaluation (RWE), to rapidly estimate the performance of network architectures. By just training the last linear classification layer, RWE reduces the computational cost of evaluating an architecture from hours to seconds. When integrated within an evolutionary multi-objective algorithm, RWE obtains a set of efficient architectures with state-of-the-art performance on CIFAR-10 with less than two hours' searching on a single GPU card. Ablation studies on rank-order correlations and transfer learning experiments to ImageNet have further validated the effectiveness of RWE.