Artificial neural networks parameters optimization design of experiments: An application in materials modeling

• Main Authors: Wimalin Laosiritaworn, Nantakarn Chotchaithanakorn
• Format: Journal
• Published: 2018
• Subjects: Biochemistry, Genetics and Molecular Biology; Chemistry; Materials Science; Mathematics; Physics and Astronomy
• Online Access: https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=67650323958&origin=inward; http://cmuir.cmu.ac.th/jspui/handle/6653943832/59408
• Institution: Chiang Mai University

This paper focused on the application of design of experiments to determine optimize parameters for multilayer-perceptron artificial neural network trained with back-propagation for modeling purpose. Artificial neural networks (ANNs) for modeling have been widely used in various fields because of its ability to 'learn' from examples. The accuracy of ANN model depends very much on the setting of network parameters, such as number of neurons, number of hidden layers and learning rate. Most literatures in this area suggested trial-anderror method for parameters setting which are time consuming and non economical, whereas the optimal setting cannot be guaranteed. Consequently, design of experiment techniques is generally required to optimize various processes. In this paper, as a case study, it was used to find optimum setting of ANN trained to model ferromagnetic material data. Interested characteristic was finite-sized ferromagnetic Curie temperature obtained from Monte Carlo simulation on two dimensional Ising spins. The results indicated that design of experiments is a promising solution to the mentioned problem. The issues arising from this case were also discussed.