Evaluating Model Performance with Hard-Swish Activation Function Adjustments

• DOI: 10.48550/arxiv.2410.06879
• Publication Date: 2024-10-09
• Authors: Sai Abhinav Pydim...; S. M. Khairnar; Sofia Garces Pala...; Ganesh Sankaranar...; Darian Hoagland; Dmitry Nepomnayshy; Huu Phong Nguyen
• Keywords: FOS: Computer and information sciences; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition

In the field of pattern recognition, achieving high accuracy is essential. While training a model to recognize different complex images, it vital fine-tune achieve highest possible. One strategy for fine-tuning involves changing its activation function. Most pre-trained models use ReLU as their default function, but switching function like Hard-Swish could be beneficial. This study evaluates performance using ReLU, Swish and functions across diverse image datasets. Our results show 2.06% increase in on CIFAR-10 dataset 0.30% ATLAS dataset. Modifying architecture lead improved overall accuracy.