Название: Neural Networks and Learning Algorithms in MATLAB
Автор: Ardahir Mohammadazadeh, Mohammad Hosein Sabzalian, Oscar Castillo
Издательство: Springer
Год: 2022
Страниц: 123
Язык: английский
Формат: pdf (true), epub
Размер: 45.2 MB

This book explains the basic concepts, theory and applications of neural networks in a simple unified approach with clear examples and simulations in the MATLAB programming language. The scripts herein are coded for general purposes to be easily extended to a variety of problems in different areas of application. They are vectorized and optimized to run faster and be applicable to high-dimensional engineering problems. This book will serve as a main reference for graduate and undergraduate courses in neural networks and applications. This book will also serve as a main basis for researchers dealing with complex problems that require neural networks for finding good solutions in areas, such as time series prediction, intelligent control and identification. In addition, the problem of designing neural network by using metaheuristics, such as the genetic algorithms and particle swarm optimization, with one objective and with multiple objectives, is presented.

Artificial neural networks are now extensively studied in order to achieve human-like efficiency. These networks consist of some linear and nonlinear computational elements that operate in tandem. Neural networks are cutting-edge computational systems and methods for Machine Learning, knowledge representation, and, finally, the application of acquired knowledge to predict outputs from complex systems. The main concept behind these networks is (to some extent) inspired by how the biological neural system processes data and information in order to learn and create knowledge. The main component of this concept is the development of novel structures for information processing systems. This system consists of many extremely interconnected processing elements known as the neurons that cooperate to solve problems and transfer information via synapses (electromagnetic communications). If a cell is damaged in these networks, other cells can compensate for its absence and contribute to its reconstruction. These networks are capable of learning. For instance, by applying burn to touch nerve cells, the cells learn not to approach hot objects, and the system learns to correct its mistake thanks to the algorithm. These systems learn comparatively; in other words, a new input is provided, and the weights of synapses change in a way that the system can generate accurate responses.

There is no agreement among researchers on how to define a neural network; however, most agree that it consists of a network of simple processing elements (neurons) capable of displaying an overall complex behavior determined by the relationship between processing elements and element parameters. The main and inspiring source for this technique is to test the central nervous system and neurons (axons, multiple branches of nerve cells, and junctions of two nerves), which are among the most important components of nervous system information processing. Simple nodes (processing elements) or units are interlinked to form a network of nodes in a neural network model. This is why they are referred to as “neural networks.” Although a neural network should not be adaptable in and of itself, it can be used practically thanks to certain algorithms designed to change the communication weight in a network (to create the desired signal).