Optimal Hyperparameters and Structure Setting of Multi-Objective Robust CNN Systems via Generalized Taguchi Method and Objective Vector Norm

TL;DR

This paper introduces a generalized Taguchi method to optimize hyperparameters and structures of multi-objective CNN systems, demonstrated on ResNet for CIFAR-10, enhancing accuracy and robustness.

Contribution

It presents a novel generalized Taguchi approach for multi-objective hyperparameter and structure optimization in CNNs, addressing robustness and performance.

Findings

Achieved optimal accuracy with ResNet on CIFAR-10

Effective multi-objective hyperparameter optimization

Validated robustness and performance improvements

Abstract

Recently, Machine Learning (ML), Artificial Intelligence (AI), and Convolutional Neural Network (CNN) have made huge progress with broad applications, where their systems have deep learning structures and a large number of hyperparameters that determine the quality and performance of the CNNs and AI systems. These systems may have multi-objective ML and AI performance needs. There is a key requirement to find the optimal hyperparameters and structures for multi-objective robust optimal CNN systems. This paper proposes a generalized Taguchi approach to effectively determine the optimal hyperparameters and structure for the multi-objective robust optimal CNN systems via their objective performance vector norm. The proposed approach and methods are applied to a CNN classification system with the original ResNet for CIFAR-10 dataset as a demonstration and validation, which shows the proposed methods are highly effective to achieve an optimal accuracy rate of the original ResNet on CIFAR-10.