Artificial Neural Networks

Abstract

Artificial neural network as the most famous artificial intelligence models are a collection of neurons with specific architecture formed based on the relationship between neurons in different layers. Neuron is a mathematical unit, and an artificial neural network that consists of neurons is a complex and nonlinear system. Artificial neural networks (ANNs) may have different architectures which result in different types of ANNs. A static ANN known as a multilayer perceptron (MLP) is the most applied ANN in different fields of engineering. This type of ANN is presented in this chapter and details on its calibration and validation are discussed. Furthermore, dynamic ANNs improved to consider the temporal dimension of data through the modeling process is presented. In this chapter, dynamic ANNs including input delay networks, recurrent networks, and a combination of both are discussed in details. Statistical neural networks, namely, radial basis estimator, generalized neural network, and probabilistic neural network, which are all developed based on a statistical-based estimation, are the third type of ANNs presented in this chapter. How to deal with calibration and validation of all models by MATLAB codes and commands are discussed. Application of the models in function approximation and data classification are presented through different examples.