Otsesuunatud tehisnärvivõrgud paketis R

Abstract

Human brain is a complex and powerful system that is able to solve a wide variety of tasks. The aim of many scientists is to develop a computer simulation that mimics the brain functions and solves problems the way our brains do. Very simplified models of biological neural networks are artificial neural networks. There are two different types of artificial neural networks – feed forward neural networks and recurrent neural networks. This thesis gives an overview of feed-forward neural networks and their working principles. The thesis is divided into two main parts. The first part is the theory of feed-forward neural networks and the second part is a practical example of neural network with software R. The first part gives an overview of the artificial neuron and its history. Also different types of artificial neurons are introduced. The first part includes instructions of how feed-forward neural networks are composed and explains how they calculate the results. Separate chapter is devoted to training artificial neural networks. The chapter gives an overview of two main training algorithms – perceptron training algorithm and back-propagation algorithm. The first is designed to train perceptrons and the second is often used in training multi-layer feed-forward neural networks. The last topic explains how to construct feed-forward neural networks with software R. It includes a tutorial of how to build a neural network that calculates the square root. The tutorial will produce a neural network which takes a single input and produces a single output. Input is the number that we want square rooting and the output is the square root of the input.