EENA: Efficient Evolution of Neural Architecture

TL;DR

EENA introduces an efficient evolutionary algorithm for neural architecture search that reduces computational costs significantly while discovering high-performing models transferable across datasets.

Contribution

The paper presents a novel mutation and crossover strategy guiding evolution with learned information, enabling faster and resource-efficient neural architecture search.

Findings

Achieves 2.56% test error on CIFAR-10 with 0.65 GPU-days

Designs architectures with 8.47M parameters that are highly effective

Discovered architectures are transferable to CIFAR-100

Abstract

Latest algorithms for automatic neural architecture search perform remarkable but are basically directionless in search space and computational expensive in training of every intermediate architecture. In this paper, we propose a method for efficient architecture search called EENA (Efficient Evolution of Neural Architecture). Due to the elaborately designed mutation and crossover operations, the evolution process can be guided by the information have already been learned. Therefore, less computational effort will be required while the searching and training time can be reduced significantly. On CIFAR-10 classification, EENA using minimal computational resources (0.65 GPU-days) can design highly effective neural architecture which achieves 2.56% test error with 8.47M parameters. Furthermore, the best architecture discovered is also transferable for CIFAR-100.