2026-05-12T23:05:02.433Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:9194 2024-12-03T20:10:47Z 3

Sine Cosine Algorithm with Tangent Search for Neural Networks Dropout Regularization, Chapter in AIS Algorithms for Intelligent Systems: ICDICI 2022: Data Intelligence and Cognitive Informatics, Springer 2022 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/9194 https://link.springer.com/chapter/10.1007/978-981-19-6004-8_59 L. Jovanovic M. Antonijevic M. Zivkovic D. Jovanovic M. Marjanovic N. Bacanin Convolutional neural networks belong to the group of deep learning methods, largely influenced by the structure and functioning of the human brain. Their primary usage is to perform image classifying tasks. All neural networks, including convolutional neural networks, are susceptible to the overfitting issue, which can occur when the network exhibits good performance on the train data while failing to accurately predict the new data when it is fed to the inputs. Few regularization approaches exist that may help in avoiding the overfit. This paper proposes a novel swarm intelligence optimization method to address the overfitting problem by choosing the adequate dropout parameter value. Swarm algorithms have previously been used to optimize the structure of neural networks; however, the full potential of these algorithms has yet not been thoroughly investigated. Scientists must invest a lot of time to choose the appropriate dropout value if they execute this task manually. In this research, an automated framework, that uses improved sine cosine metaheuristics to perform this task, is proposed. The proposed framework was tested on four standard benchmark datasets, namely MNIST, CIFAR10, Semeion, and UPS. The simulation results have been validated against the results generated by several other state-of-the-art swarm intelligence algorithms. The comparison shows that the proposed method outperforms other cutting-edge algorithms in terms of classification error, therefore achieving higher accuracy percentage. bookPart Springer, Singapore 789 802 10.1007/978-981-19-6004-8_59 AIS Algorithms for Intelligent Systems: ICDICI 2022: Data Intelligence and Cognitive Informatics