Probabilistic Reasoning and Bayesian Belief Networks / Neural Networks / Applications of Modern Heuristic Methods

This three-volume set, "Probabilistic Reasoning and Bayesian Belief Networks", "Neural Networks" and "Applications of Modern Heuristic Methods", which individually stand alone, but combined form a set treating a broad but integrated spectrum of techniques and tools for undertaking complex tasks. Volume one: one of the most significant characteristics of an intelligent computer system is the ability to reason with judgmental knowledge. That is, how it uses heuristics, and improves its decision-making procedures in the light of examples which it is given. These heuristics are typically uncertain. Numerous methods have been suggested and are used for dealing with uncertainty. Many have been developed to overcome particular problems associated with the use of classical formalism for dealing with uncertainty, for example, probability theory. Recent work in theoretical statistics has demonstrated that it is possible to adopt a sound probabilistic approach to uncertain inference using Bayesian belief networks - a graphical representation of causal dependencies. This book summarizes some important work in the development of computational models of Bayesian belief networks, and their applications to medicine, transport and defence. Volume two: neural networks, based on simple adaptive models of living neurons, has shown itself able to tackle a very wide range of problems, particularly in conjunction with probabilistic reasoning and Bayesian belief networks using local computations. Similarly genetic algorithms and simulated annealing have been used for the solution of a large range of optimization problems. Complex tasks in areas such as machine vision, time-series analysis, robotics, control, cost analysis, and even share price and currency prediction are routinely handled by neural networks. They are also proving of importance in the development of mixed hybrid systems, which utilize other approaches to hard information processing techniques. Genetic algorithms, expert systems and Bayesian techniques are all being combined with neural network constructions to give ever-more powerful applications. Volume three: many problems which arise in industry and commerce, including scheduling, strategic planning, routing and location prove extremely difficult to solve optimally. The mathematical models produced may require an unacceptable timescale to solve on even the most powerful of modern super computers. In practice, it is only feasible to use some method which produces a good, but not necessarity optimal solution. Heuristic methods exploit techniques which are intuitively reasonable, and which, by experience, can be shown to work acceptably well. Some modern heuristic methods exploit recent developments in artificial intelligence and, to some extent, attempt to mimic natural processes. This is the case for tabu search, simulated annealing, genetic algorithms and their hybrids. These methods provide fast and effective ways of solving some of the most important and difficult problems arising in industry. There is a skill to be acquired if they are to be used to best effect and this book, by covering a number of case studies, attempts to impart this skill. The book should be of interest to all those working in: adaptive information processing, particularly in the allied fields of computer science, electrical engineering, physics and mathematics; also those researching in the neurosciences and branches of psychology and philsophy, particularly those concerned with neural modelling should benefit from this book. Corporate users should include IT specialists , production and control engineers, research and development departments, and consultants.

Table of Contents:
Volume one: from Bayesian netorks to causal networks; exact and approximate algorithms and their implementations in mixed graphical models; models and modelling in context; modelling ignorance in uncertainty theories; choosing network complexity; a system for hypothesis-driven data request; an efficient graphical algorithm for updating the estimates of the dispersal of gaseous waste after an accidental release; graphical representation of a network traffic model; a C++-class library for building Bayesian belief networks; smoothing noisy signals with Bayesian networks; efficient multiple-disorder diagnosis by strategic focusing; weighted inference rules and Bayesian belief networks; on the idiot vs proper Bayes approach in clinical diagnostic systems; constructing computationally-efficient Bayesian models via unsupervised clustering; Bayesian graphical models of the natural history of HIV-infection. Volume two: the promise of neural networks; neural network applications - some case studies; applying neural networks in real applications; neural computing awareness and technology transfer; the theory and practice of N-tuple neural networks; performing variable binding with a neural network; neural net training - random versus systematic; segmentation and matching in infra-red airborne images using a binary neural network; optic, flow determination by means of a neural net; neural networks and data visualization; the use of neural networks for region labelling and scene understanding; real-time control of a high-temperature plasma using a hardware neural network; route-finding by neural nets; a comparison of traditional methods, statistical techniques and neural networks for machine-condition monitoring; measuring the performance of neural networks in modern portfolio management - testing strategies and metrics; neural networks - managing innovations; building hybrid systems with neural networks and genetic algorithms; neuro-fuzzy high-dimensional approximation; neuro-fuzzy networks for process modelling and fault diagnosis; hardware-realizable neural networks; modelling consciousness. Volume three: Part 1 Techniques and principles: variants of simulated annealing for practical problem-solving; a unified approach to tabu search, simulated annealing and genetic algorithms; designing adaptable systems through the study and application of biological sources; a review of models for simple and cellular genetic algorithms. Part 2 Applications: economic applications of genetic algorithms; genetics-based machine learning - applications in industry and commerce; genetic algorithms and combinatorial optimization. Part 3 Case studies: studies of continuous-flow chemical synthesis using genetical algorithms; getting the timing right; the use of genetic algorithms in scheduling; location of concentrators using simulated annealing; a simulated annealing post-processor for the vehicle-routing problem. Part 4 Toolkits: the GAmeter toolkit.