Automating Neural Architecture Design without Search

TL;DR

This paper introduces a novel method for neural architecture design that bypasses traditional search processes by learning from expert-designed architectures, significantly reducing computational costs while maintaining high performance.

Contribution

The paper proposes a knowledge-based approach using graph neural networks to directly generate neural architectures, eliminating the need for costly search and evaluation phases.

Findings

Achieved 97.82% accuracy on CIFAR10 with minimal computational cost.

Demonstrated high transferability of the learned knowledge across different search spaces.

Reduced architecture design cost by orders of magnitude compared to traditional NAS methods.

Abstract

Neural structure search (NAS), as the mainstream approach to automate deep neural architecture design, has achieved much success in recent years. However, the performance estimation component adhering to NAS is often prohibitively costly, which leads to the enormous computational demand. Though a large number of efforts have been dedicated to alleviating this pain point, no consensus has been made yet on which is optimal. In this paper, we study the automated architecture design from a new perspective that eliminates the need to sequentially evaluate each neural architecture generated during algorithm execution. Specifically, the proposed approach is built by learning the knowledge of high-level experts in designing state-of-the-art architectures, and then the new architecture is directly generated upon the knowledge learned. We implemented the proposed approach by using a graph neural network for link prediction and acquired the knowledge from NAS-Bench-101. Compared to existing peer competitors, we found a competitive network with minimal cost. In addition, we also utilized the learned knowledge from NAS-Bench-101 to automate architecture design in the DARTS search space, and achieved 97.82% accuracy on CIFAR10, and 76.51% top-1 accuracy on ImageNet consuming only $2\times10^{-4}$ GPU days. This also demonstrates the high transferability of the proposed approach, and can potentially lead to a new, more computationally efficient paradigm in this research direction.