Rethinking the Number of Channels for the Convolutional Neural Network

TL;DR

This paper introduces a fast, resource-efficient method for automatically determining the optimal number of channels in convolutional neural networks, improving accuracy and reducing parameters on CIFAR datasets.

Contribution

It presents a novel functionally incremental search method focused on network widths, enabling rapid and effective channel configuration without extensive computational resources.

Findings

Achieved about 0.5% accuracy improvement on CIFAR-10

Achieved about 2.33% accuracy improvement on CIFAR-100

Required only 0.4 to 1.3 GPU-days for search

Abstract

Latest algorithms for automatic neural architecture search perform remarkable but few of them can effectively design the number of channels for convolutional neural networks and consume less computational efforts. In this paper, we propose a method for efficient automatic architecture search which is special to the widths of networks instead of the connections of neural architecture. Our method, functionally incremental search based on function-preserving, will explore the number of channels rapidly while controlling the number of parameters of the target network. On CIFAR-10 and CIFAR-100 classification, our method using minimal computational resources (0.4~1.3 GPU-days) can discover more efficient rules of the widths of networks to improve the accuracy by about 0.5% on CIFAR-10 and a~2.33% on CIFAR-100 with fewer number of parameters. In particular, our method is suitable for exploring the number of channels of almost any convolutional neural network rapidly.