Machine Intelligence and Pattern Recognition Series

Machine Intelligence and Pattern Recognition, Volume 2: Computational Geometry focuses on the operations, processes, methodologies, and approaches involved in computational geometry, including algorithms, polygons, convex hulls, and bucketing techniques. The selection first ponders on optimal parallel algorithms for selection, sorting, and computing convex hulls, simple on-line algorithms for convex polygons, and geometric algorithms that use the furthest-point Voronoi diagram. Discussions focus on algorithms that use the furthest-point Voronoi diagram, intersection of a convex polygon and a halfplane, point insertion, convex hulls and polygons and their representations, and parallel algorithm for selection and computing convex hulls. The text then examines optimal convex decompositions, expected time analysis of algorithms in computational geometry, and practical use of bucketing techniques in computational geometry. The book takes a look at minimum decompositions of polygonal objects, framework for computational morphology, display of visible edges of a set of convex polygons, and implementation study of two algorithms for the minimum spanning circle problem. Topics include rolling algorithm, shape of point sets, and decomposition of rectilinear and simple polygons and polygons with holes. The selection is a valuable source of data for researchers interested in computational geometry.

Techniques for 3-D Machine Perception

Computational Geometry is a new discipline of computer science that deals with the design and analysis of algorithms for solving geometric problems. There are many areas of study in different disciplines which, while being of a geometric nature, have as their main component the extraction of a description of the shape or form of the input data. This notion is more imprecise and subjective than pure geometry. Such fields include cluster analysis in statistics, computer vision and pattern recognition, and the measurement of form and form-change in such areas as stereology and developmental biology. This volume is concerned with a new approach to the study of shape and form in these areas. Computational morphology is thus concerned with the treatment of morphology from the computational geometry point of view. This point of view is more formal, elegant, procedure-oriented, and clear than many previous approaches to the problem and often yields algorithms that are easier to program and have lower complexity.

This volume brings together the results of research into the methodology and applications of pattern recognition, with particular emphasis given to the incorporation of artificial intelligence methodologies into pattern recognition systems. The first part of this volume covers image analysis and processing software, systems and algorithms. Pattern analysis and classifier design are dealt with in part two, while the last part deals with model based and expert systems, including uncertainty calculus methods in pattern analysis and object recognition. A number of specific application areas are considered, including such diverse topics as fingerprinting, astronomy, molecular biology and pathology.

How to deal with uncertainty is a subject of much controversy in Artificial Intelligence. This volume brings together a wide range of perspectives on uncertainty, many of the contributors being the principal proponents in the controversy. Some of the notable issues which emerge from these papers revolve around an interval-based calculus of uncertainty, the Dempster-Shafer Theory, and probability as the best numeric model for uncertainty. There remain strong dissenting opinions not only about probability but even about the utility of any numeric method in this context.

Clearly illustrated in this volume is the current relationship between Uncertainty and AI. It has been said that research in AI revolves around five basic questions asked relative to some particular domain: What knowledge is required? How can this knowledge be acquired? How can it be represented in a system? How should this knowledge be manipulated in order to provide intelligent behavior? How can the behavior be explained? In this volume, all of these questions are addressed. From the perspective of the relationship of uncertainty to the basic questions of AI, the book divides naturally into four sections which highlight both the strengths and weaknesses of the current state of the relationship between Uncertainty and AI.

This second volume is arranged in four sections: Analysis contains papers which compare the attributes of various approaches to uncertainty. Tools provides sufficient information for the reader to implement uncertainty calculations. Papers in the Theory section explain various approaches to uncertainty. The Applications section describes the difficulties involved in, and the results produced by, incorporating uncertainty into actual systems.

This volume, like its predecessors, reflects the cutting edge of research on the automation of reasoning under uncertainty. A more pragmatic emphasis is evident, for although some papers address fundamental issues, the majority address practical issues. Topics include the relations between alternative formalisms (including possibilistic reasoning), Dempster-Shafer belief functions, non-monotonic reasoning, Bayesian and decision theoretic schemes, and new inference techniques for belief nets. New techniques are applied to important problems in medicine, vision, robotics, and natural language understanding.

With the growing complexity of pattern recognition related problems being solved using Artificial Neural Networks, many ANN researchers are grappling with design issues such as the size of the network, the number of training patterns, and performance assessment and bounds. These researchers are continually rediscovering that many learning procedures lack the scaling property; the procedures simply fail, or yield unsatisfactory results when applied to problems of bigger size. Phenomena like these are very familiar to researchers in statistical pattern recognition (SPR), where the curse of dimensionality is a well-known dilemma. Issues related to the training and test sample sizes, feature space dimensionality, and the discriminatory power of different classifier types have all been extensively studied in the SPR literature. It appears however that many ANN researchers looking at pattern recognition problems are not aware of the ties between their field and SPR, and are therefore unable to successfully exploit work that has already been done in SPR. Similarly, many pattern recognition and computer vision researchers do not realize the potential of the ANN approach to solve problems such as feature extraction, segmentation, and object recognition. The present volume is designed as a contribution to the greater interaction between the ANN and SPR research communities.

Parallel processing for AI problems is of great current interest because of its potential for alleviating the computational demands of AI procedures. The articles in this book consider parallel processing for problems in several areas of artificial intelligence: image processing, knowledge representation in semantic networks, production rules, mechanization of logic, constraint satisfaction, parsing of natural language, data filtering and data mining. The publication is divided into six sections. The first addresses parallel computing for processing and understanding images. The second discusses parallel processing for semantic networks, which are widely used means for representing knowledge - methods which enable efficient and flexible processing of semantic networks are expected to have high utility for building large-scale knowledge-based systems. The third section explores the automatic parallel execution of production systems, which are used extensively in building rule-based expert systems - systems containing large numbers of rules are slow to execute and can significantly benefit from automatic parallel execution. The exploitation of parallelism for the mechanization of logic is dealt with in the fourth section. While sequential control aspects pose problems for the parallelization of production systems, logic has a purely declarative interpretation which does not demand a particular evaluation strategy. In this area, therefore, very large search spaces provide significant potential for parallelism. In particular, this is true for automated theorem proving. The fifth section considers the problem of constraint satisfaction, which is a useful abstraction of a number of important problems in AI and other fields of computer science. It also discusses the technique of consistent labeling as a preprocessing step in the constraint satisfaction problem. Section VI consists of two articles, each on a different, important topic. The first discusses parallel formulation for the Tree Adjoining Grammar (TAG), which is a powerful formalism for describing natural languages. The second examines the suitability of a parallel programming paradigm called Linda, for solving problems in artificial intelligence. Each of the areas discussed in the book holds many open problems, but it is believed that parallel processing will form a key ingredient in achieving at least partial solutions. It is hoped that the contributions, sourced from experts around the world, will inspire readers to take on these challenging areas of inquiry.

With the increasing availability of parallel machines and the raising of interest in large scale and real world applications, research on parallel processing for Artificial Intelligence (AI) is gaining greater importance in the computer science environment. Many applications have been implemented and delivered but the field is still considered to be in its infancy. This book assembles diverse aspects of research in the area, providing an overview of the current state of technology. It also aims to promote further growth across the discipline. Contributions have been grouped according to their subject: architectures (3 papers), languages (4 papers), general algorithms (6 papers), and applications (5 papers). The internationally sourced papers range from purely theoretical work, simulation studies, algorithm and architecture proposals, to implemented systems and their experimental evaluation. Since the book is a second volume in the parallel processing for AI series, it provides a continued documentation of the research and advances made in the field. The editors hope that it will inspire readers to investigate the possiblities for enhancing AI systems by parallel processing and to make new discoveries of their own!

The era of detailed comparisons of the merits of techniques of pattern recognition and artificial intelligence and of the integration of such techniques into flexible and powerful systems has begun. So confirm the editors of this fourth volume of Pattern Recognition in Practice, in their preface to the book. The 42 quality papers are sourced from a broad range of international specialists involved in developing pattern recognition methodologies and those using pattern recognition techniques in their professional work. The publication is divided into six sections: Pattern Recognition, Signal and Image Processing, Probabilistic Reasoning, Neural Networks, Comparative Studies, and Hybrid Systems, giving prospective users a feeling for the applicability of the various methods in their particular field of specialization.