Analyze and Design Network Architectures by Recursion Formulas

TL;DR

This paper introduces a mathematical formula-based methodology for designing neural network architectures, demonstrating significant performance improvements through experiments on CIFAR and ImageNet datasets.

Contribution

It proposes a novel approach to architecture design using recursion formulas, enabling systematic creation of improved neural networks.

Findings

New architectures outperform ResNet on CIFAR and ImageNet

Recursion formulas effectively capture differences in network designs

Significant performance gains achieved through the proposed method

Abstract

The effectiveness of shortcut/skip-connection has been widely verified, which inspires massive explorations on neural architecture design. This work attempts to find an effective way to design new network architectures. It is discovered that the main difference between network architectures can be reflected in their recursion formulas. Based on this, a methodology is proposed to design novel network architectures from the perspective of mathematical formulas. Afterwards, a case study is provided to generate an improved architecture based on ResNet. Furthermore, the new architecture is compared with ResNet and then tested on ResNet-based networks. Massive experiments are conducted on CIFAR and ImageNet, which witnesses the significant performance improvements provided by the architecture.