Название: Computational Intelligence and Mathematics for Tackling Complex Problems 3
Автор: Istvan A. Harmati, Laszlo T. Koczy, Jesus Medina
Издательство: Springer
Серия: Studies in Computational Intelligence
Год: 2022
Страниц: 223
Язык: английский
Формат: pdf (true)
Размер: 10.1 MB

Complex problems and systems, which prevail in the real world, cannot often be tackled and solved either by traditional methods offered by mathematics or even the traditional Computer Science (CS) and Artificial Intelligence (AI). What is the way out of this dilemma? Advanced methodologies, and tools and techniques, "mimicking" human reasoning or the behavior of animals, animal populations or certain parts of the living bod, based on traditional Computer Science and the initial approaches of Artificial Intelligence are often referred to as biologically inspired methods, or often Computational Intelligence (CI). Computational Intelligence offers effective and efficient solutions to many "unsolvable" problems. However, it is far from being a ready to use and complete collection of approaches, and is rather a continuously developing field without clear borders. The emerging new models and algorithms of Computational Intelligence are deeply rooted in the vast apparatus of traditional mathematics.

The Chapter 20 refers to the computational intelligence methods in particular to theory of artificial neural networks. The paper proposes three algorithms for training a general regression neural network. The first uses a nature-inspired optimization approach known as particle swarm optimization, while the latter two, i.e., the plug-in and the cross-validation, are based on classical mathematical methods, including the theory of kernel density estimators. The aforementioned algorithms are applied to determining network smoothing parameters - which is the main task in GRNN learning. The GRNN constitutes extension of probabilistic neural network structure. In this work, the trained GRNN undergoes benchmarking on repository data sets.

Computational intelligence methods are used very widely in tasks such as soft computing (consisting of approximating the value of solving a complex problem with an assumed level of accuracy), through the application of fuzzy logic and biologically inspired methods, and ending with simulating human intelligence. Artificial neural networks are one of the tools of computational intelligence. These are structures inspired by biological neural networks that make up the brain.