Regularized Evolution for Image Classifier Architecture Search

By Esteban Real et al

Read the original document by opening this link in a new tab.

Table of Contents

Abstract
Introduction
Related Work
Methods
Baseline Algorithms
Experimental Setup
Methods Details

Summary

The document discusses the use of regularized evolution for image classifier architecture search, presenting the evolution of an image classifier called AmoebaNet-A that surpasses hand-designed models. It introduces modifications to the evolutionary algorithm, including aging evolution and specific mutations in the NASNet search space. The paper compares the evolutionary method to reinforcement learning and random search, demonstrating the effectiveness of evolution in discovering high-quality architectures.