Optimizing Hyperparameters for Enhanced Performance in Convolutional Neural Networks: A Study Using NASNetMobile and DenseNet201 Models

Author:

Aksoy İbrahim1ORCID,Adem Kemal2ORCID

Affiliation:

1. AKSARAY UNIVERSITY

2. SİVAS BİLİM VE TEKNOLOJİ ÜNİVERSİTESİ, MÜHENDİSLİK VE DOĞA BİLİMLERİ FAKÜLTESİ, BİLGİSAYAR MÜHENDİSLİĞİ BÖLÜMÜ

Abstract

Convolutional neural networks, inspired by the workings of biological neural networks, have proven highly successful in tasks like image data recognition, classification, and feature extraction. Yet, designing and implementing these networks pose certain challenges. One such challenge involves optimizing hyperparameters tailored to the specific model, dataset, and hardware. This study delved into how various hyperparameters impact the classification performance of convolutional neural network models. The investigation focused on parameters like the number of epochs, neurons, batch size, activation functions, optimization algorithms, and learning rate. Using the Keras library, experiments were conducted using NASNetMobile and DenseNet201 models—highlighted for their superior performance on the dataset. After running 65 different training sessions, accuracy rates saw a notable increase of 6.5% for NASNetMobile and 11.55% for DenseNet201 compared to their initial values.

Publisher

Bandirma Onyedi Eylul University

Reference33 articles.

1. E. Öztemel “Yapay sinir ağları”, Papatya Yayıncılık, İstanbul, 2003.

2. S. Aktürk and K. Serbest, “Nesne Tespiti İçin Derin Öğrenme Kütüphanelerinin İncelenmesi”, Journal of Smart Systems Research, vol. 3, no. 2, pp. 97-119, 2022.

3. A. Onan, “Evrişimli sinir ağı mimarilerine dayalı Türkçe duygu analizi”, Avrupa Bilim ve Teknoloji Dergisi, pp. 374-380, 2020.

4. L.N. Smith, “Cyclical learning rates for training neural networks”, IEEE winter conference on applications of computer vision (WACV), pp. 464-472, 2017.

5. C. Bircanoğlu and N. Arıca, “Yapay Sinir Ağlarında Aktivasyon Fonksiyonlarının Karşılaştırılması”, in 2018 26th signal processing and communications applications conference (SIU). IEEE, pp. 1-4, İzmir, 2018.

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3