MULTI-OBJECTIVE EVOLUTIONARY TRAINING SET SELECTION FOR ARTIFICIAL NEURAL NETWORKS

Abstract

Artificial Neural Networks (ANNs) attempt to simulate biological systems, corresponding to the human brain. A study of human brain reveals that it contains neurons that are interconnected to each other by nodes, and these nodes are called as synapses. In these networks of neurons, learning takes place by the change in strength of the synaptic connections which comes as a response of electrical impulses. Artificial neural networks mimics this behavior of human neural networks and hence it retains the name [5]. The basic building block of ANN is called a unit or neuron. ANNs can be in several types of architectures based on the way neurons are connected to each other. Perceptron is the basic type of ANNS, its structure is designed into a set of input nodes and an output node. [5]. Based on the architecture of the neural network and the direction of information flow, ANNs are divided into feedforward ANNs and recurrent ANNs. As the ANN model in the present research is developed using a feedforward network, a brief description for this type is present in the next section. 2.1.1 Feedforward Neural Networks (MLPs) When multiple ANN units are designed in a way that the information flows from input nodes through one or multiple hidden layers to the output node, it forms a multi-layer feedforward neural network. These are also called as multi-layer perceptrons (MLP). MLPs are the most popular types of ANNs which extensively researched and implements in real-life situations.