CONet: Channel Optimization for Convolutional Neural Networks

TL;DR

This paper introduces CONet, an efficient dynamic scaling algorithm that automatically optimizes channel sizes in CNNs, improving accuracy and efficiency across multiple datasets and architectures.

Contribution

CONet provides a novel, automatic method for channel size optimization in CNNs, overcoming limitations of discrete sampling and manual tuning.

Findings

CONet outperforms baseline models on CIFAR10/100 and ImageNet datasets.

It effectively optimizes channel sizes in ResNet, DARTS, and DARTS+ architectures.

The introduced metrics accurately identify optimal information accumulation during training.

Abstract

Neural Architecture Search (NAS) has shifted network design from using human intuition to leveraging search algorithms guided by evaluation metrics. We study channel size optimization in convolutional neural networks (CNN) and identify the role it plays in model accuracy and complexity. Current channel size selection methods are generally limited by discrete sample spaces while suffering from manual iteration and simple heuristics. To solve this, we introduce an efficient dynamic scaling algorithm -- CONet -- that automatically optimizes channel sizes across network layers for a given CNN. Two metrics -- "\textit{Rank}" and "\textit{Rank Average Slope}" -- are introduced to identify the information accumulated in training. The algorithm dynamically scales channel sizes up or down over a fixed searching phase. We conduct experiments on CIFAR10/100 and ImageNet datasets and show that CONet can find efficient and accurate architectures searched in ResNet, DARTS, and DARTS+ spaces that outperform their baseline models. This document supersedes previously published paper in ICCV2021-NeurArch workshop. An additional section is included on manual scaling of channel size in CNNs to numerically validate of the metrics used in searching optimum channel configurations in CNNs.