A Study on Layers of Deep Neural Networks

Artificial intelligence is a field of technology that can reflect the same abilities as human intelligence through machines. In recent information systems, artificial intelligence is applied in various areas to solve problems effectively. Among the various artificial intelligence technologies, deep learning is used in prediction problems through trained models from a set of previously known training data. The deep learning uses neural networks designed with multiple layers to establish a model for training to find a solution for problems. In designing a neural network, it is important to decide the structures of layers that are the basic elements for maintaining the features of a given problem identified from learning procedures with a set of input training data. Thus, deciding the number of layers and nodes in the neural network for constructing layers is one of the most basic criteria for designing a neural network. In this study, we evaluate the effect of the numbers of layers and nodes in the neural network design with respect to the training accuracy of the neural network and study the composition of layers for designing an efficient neural network.