Neural nets. Optimization by learning and evolution
Both the field of artificial neural networks and genetic algorithms belong recently to widely recognized branches of modern computer science. Both of them have found a lot of practical applications. The underlying book shows how to apply and combine the most promising ideas, models and methods from these two areas in order to achieve powerful means for solving various optimization tasks. In principle, the text can be divided into three main parts. The first part (Chapter 2) covers fundamental principles of basic neural network models with the emphasis on knowledge representation in neural networks. The described neural network models comprise the Multilayer Perceptron networks and selected learning algorithms based on gradient descent, Radial Basis Function networks, Self-Organizing Feature Maps, Adaptive Resonance Theory and Hopfield-like networks. The next part of the book (Chapter 3) is dedicated to neural models for strategy learning. Above all, it deals with reinforcement learning -- more attention is here given to Temporal Difference learning. The last part (Chapter 4) concerns the basics of the theory of genetic algorithms and the main principles of an evolutionar network optimizer ENZO (Evolutionärer Netzwerk-Optimierer) implemented by the author. In the end, this chapter reviews also general ideas of a system for parallelizing the network learning and evolution PINK (Parallel Intelligent Neural Network Simulator Karlsruhe). This book originated from the author's habilitation thesis at University Karlsruhe. His goal was to propose a software system for designing, training and optimizing neural networks by learning and evolution (ENZO). Therefore, especially those models (Multilayer Perceptron and Radial Basis Function networks) and learning algorithms (Gradient-Descent method, Temporal Difference learning and genetic algorithms) implemented within the framework of ENZO are discussed here carefully and in more detail than in most textbooks. The analysis of these paradigms includes a lot of well documented experiments and graphs. Moreover, original results achieved by the author, his colleagues and students are reported in the text, too. Nevertheless, all of the most important neural network models are described at least briefly and sufficient references are stated. The book contains several figures illustrating the ideas explained in the text. Unfortunately, Figure 4.34 on p. 263 demonstrates two separate 4-1-4 networks with different inputs working in parallel rather than a single transformed 8-2-8 network with an intrinsic parallelism. This book provides a detailed discussion of the models and learning algorithms selected for implementation within ENZO. Therefore, it might be of great value for those readers interested in implementation and application of the above models and methods. For the same reason, the text is surely recommendable also to student already familiar with the foundations and main problems connected with artificial neural networks and seeking for deeper understanding of the respective paradigms.
- Neural networks in optimization
- Neural networks and evolutionary algorithms
- Structure determination and training of neural networks using evolution programs
- scientific article; zbMATH DE number 5280204 (Why is no real title available?)
- scientific article; zbMATH DE number 1304439 (Why is no real title available?)
- scientific article; zbMATH DE number 1067617 (Why is no real title available?)
- Neural network training and simulation using a multidimensional optimization system
- scientific article; zbMATH DE number 1935427 (Why is no real title available?)
- SIMPLE NEURAL NETWORKS THAT OPTIMIZE DECISIONS
This page was built for publication: Neural nets. Optimization by learning and evolution
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4350446)
