Yuhui Shi

Implementation of evolutionary fuzzy systems

Evolutionary fuzzy systems are discussed in which the membership function shapes and types and the fuzzy rule set including the number of rules inside it are evolved using a genetic (evolutionary) algorithm. In addition, the genetic parameters (operators) of the evolutionary algorithm are adapted via a fuzzy system. Benefits of the methodology are illustrated in the process of classifying the iris data set. Possible extensions of the methods are summarized.

chapter two – Computational intelligence

Publisher Summary This chapter deals with the key elements of computational intelligence (CI): adaptation and learning. Dynamic adaptation is the ability of a system to adapt “online”—in essentially real time—in a changing environment. In dynamic adaptation, the system adapts while it is running (online), rather than being taken offline to be retrained. For a system to exhibit adaptation, its trajectory through the problem space must depend on the state of its environment. There are three categories of adaptation pertinent to computational intelligence: supervised adaptation, reinforcement adaptation, and unsupervised adaptation. Reinforcement adaptation of a system is achieved through its interaction with a “critic” that provides heuristic reinforcement information. The input variable information often includes the dynamic range of each variable and perhaps other variable information such as the precision required. In the case of unsupervised adaptation, no external teacher or critic is involved in system adaptation. Instead, a dataset comprising example vectors of the systems variable parameters is provided, which is operated on by the unsupervised learning algorithm.

Using artificial neural network for sleep/wake discrimination from wrist activity: preliminary results

In this paper, we investigate the possibility of using artificial neural networks for discrimination of sleep from wakefulness. Several feature extraction methods are employed. The preliminary results show that the artificial neural networks are good candidates for the task.

Neural network concepts and paradigms

This chapter provides an overview of neural networks. Neural networks consist of processing elements and weighted connections. Each layer in a neural network consists of a collection of processing elements. Each processing element (PE) collects the values from all of its input connections, performs a predefined mathematical operation, and produces a single output value. The combination of processing elements and weighted connections creates a neural network topology. A convenient analogy is a directed graph, where the edges are analogous to the connection weights and the nodes are analogous to the processing elements. Neural networks cannot operate without data. Some neural networks use only single patterns and others use pattern pairs. The dimensionality of the input pattern is not necessarily the same as the output pattern. When a network uses only single patterns, it is defined as an autoassociative network, and when a network uses pattern pairs, it is heteroassociative.

Review of biomedical applications of computational intelligence

Computational intelligence consists of artificial neural networks, evolutionary computation, fuzzy logic systems and the combination of these three components with each other and/or with other traditional approaches. In the past several years, there has been a phenomenal growth in the research and development of technology of computational intelligence. Computational intelligence has found applications in many areas; among them biomedical engineering has evolved into one of the major application areas of computational intelligence. In this paper, we give a brief review of biomedical applications of computational intelligence.

Evolutionary computation concepts and paradigms

This chapter provides an introduction to one of the component methodologies of computational intelligence—evolutionary computation (EC). One of the areas of evolutionary computation is evolutionary programing, which uses the selection of the fittest, but the only structure-modifying operation allowed is mutation. Another area of evolutionary computation is particle swarm optimization, which has roots in three main component areas: artificial life, evolutionary computation, and social psychology. Evolutionary computation paradigms do not require information that is auxiliary to the problem, such as function derivatives. In EC optimization paradigms, the fitness of each member of the population is calculated from the value of the function being optimized, and it is common to use the function output as the measure of fitness. Some EC paradigms, particularly genetic algorithms, use special encodings for the parameters of the problem being solved. In genetic algorithms, the parameters are often encoded as binary strings, but any finite alphabet can be used.

chapter ten – Performance metrics

Publisher Summary This chapter examines some issues related to the performance of a computational intelligence implementation. Some of the general issues are specifying the sizes and numbers of iterations for training datasets and the selection of test datasets for neural networks. Other issues are the selection of the “gold standards” against which performance is measured and the role the decision threshold level of a processing element in a neural network can play in determining system performance. Two issues are associated with the selection of gold standards, for both training sets and testing sets: the classification itself and the selection of a representative pattern set. When training some neural networks, especially back-propagation networks, it is often a good idea to select a training set with about the same number of patterns for each classification. When the network has three output processing elements (PEs), each of which becomes active for a particular pattern classification, it is a good idea to have a training pattern set with about one-third of the patterns from each classification.

Fuzzy systems implementations

This chapter presents two implementations of fuzzy systems—fuzzy rule systems and evolutionary fuzzy rule systems. The implementation of fuzzy rule systems and all other implementations are written in C++. In C, a struct data structure is defined to include all the related data and even some methods (functions); in C++, a new class is defined that binds the data and methods together. The fuzzy rule system implementation is a flexible tool that is capable of solving a wide variety of classification and diagnostic problems. It utilizes user-defined triangular and/or nonlinear membership functions. The executable code for the system is in the file f l.exe , and the specifications of files and other parameters appear in a run file, filename.run . To run the system, at the system prompt type fl filename.run , making sure that the run file is in the same directory as the executable.

Fuzzy systems concepts and paradigms

This chapter presents the computational intelligence component methodology that is known as “fuzzy logic.” Fuzzy logic comprises fuzzy sets, which are a way of representing nonstatistical uncertainty, and approximate reasoning, which includes the operations used to make inferences in fuzzy logic. There are many ways to represent degrees of membership in fuzzy sets. The most common is a representation of the form μ A ( x ) = m , which states that the membership value of x in the fuzzy set A is m , where 0 ≤ m ≤ 1. A fuzzy set on a numeric variable such as height or temperature is represented by a fuzzy membership function. The function can be linear, either descending or ascending; normal, bell-shaped, or triangular; an S-shaped (sigmoid or logistic) function; or arbitrary or irregular or it can have plateaus or “shoulders.”

chapter one – Foundations

Publisher Summary This chapter provides an introduction to component methodologies—computational intelligence (CI), including artificial neural networks, fuzzy logic, and evolutionary computation. Computational intelligence systems usually incorporate hybrids of paradigms such as artificial neural networks, fuzzy systems, and evolutionary computation systems, augmented with knowledge elements. They are often designed to mimic one or more aspects of biological intelligence. The chapter reviews the biological bases for artificial neural network and evolutionary computation analysis tools. The concept of chromosomes is central to both genetics and evolutionary computation. Individual patterns, or strings, in evolutionary computation systems are basically analogous to chromosomes in biological systems. In the artificial chromosomes of evolutionary computation systems, the chromosome patterns or strings are made up of parameters, or features, that can vary over a specified range of values. A given parameter or feature occupies a fixed location in the artificial chromosome. The chromosome therefore is encoded to represent a set of parameters.