Search

The aim of this work is to search for a Convolutional Neural Network (CNN) architecture that performs optimally across all factors, including accuracy, memory footprint, and computing time, suitable for mobile devices. Although deep learning has evolved for use on devices with minimal resources, its implementation is hampered by that these devices are not designed to tackle complex tasks, such as CNN architectures. To address this limitation, a Network Architecture Search (NAS) strategy is considered, which employs a Multi-Objective Evolutionary Algorithm (MOEA) to create an efficient and robust CNN architecture by focusing on three objectives: fast processing times, reduced storage, and high accuracy. Furthermore, we proposed a new Efficient CNN Population Initialization (ECNN-PI) method that utilizes a combination of random and selected strong models to generate the first-generation population. To validate the proposed method, CNN models are trained using CIFAR-10, CIFAR-100, ImageNet, STL-10, FOOD-101, THFOOD-50, FGVC Aircraft, DTD, and Oxford-IIIT Pets benchmark datasets. The MOEA-Net algorithm outperformed other models on CIFAR-10, whereas MOEANet with the ECNN-PI method outperformed other models on CIFAR-10 and CIFAR-100. Furthermore, both the MOEA-Net algorithm and MOEA-Net with the ECNN-PI method outperformed DARTS, P-DARTS, and Relative-NAS for small-scale multi-class and fine-grained datasets. [ABSTRACT FROM AUTHOR]

NUF-Net (Naresuan University and Fiber One Public Company Limited Network) is a new and improved Convolutional Neural Network (CNN) model based on the previously developed NU-LiteNet model. Improvements in accuracy were achieved by adding the identity mapping technique of the ResNet model and incorporating Snapshot Ensembles and the Cutout technique into the NU-LiteNet model. We modified the structure of the convolution layers by changing any filters of a size larger than 3 ×3, into a 3 ×3 filter, thereby significantly reducing processing time and reducing the error rate. To test the effectiveness of our modifications, we developed 10 variations of the NUF-Net-Residual model, one of which, termed NUF-Net-Residual-102, achieved significantly lower error rates than both ResNet and Wide-ResNet when using CIFAR-10, CIFAR-100 and Tiny-ImageNet datasets. The relative error rates were 2.94% for CIFAR-10, 17.57% for CIFAR-100 and 29.57% for Tiny-ImageNet. As well, NUF-Net-Residual-102 achieved a model parameter size of 31.65 million which is a lower value than for Wide-ResNet-32 (46.16 million), although higher than ResNet-1202 (19.42 million). [ABSTRACT FROM AUTHOR]