Building Expert Systems is an excellent text in probably the hottest area of artificial intelligence, if not in software in general. The book is the result of an expert ...
network sensitivity calculations using Tellegen’s theorem, broadband impedance matching, linear amplifier design (using, for instance, Smith charts, scattering parameters, and transducer gains), and elementary filter design. In each case, the computer programs given in the text can be used for part or all of the synthesis work. The author states that the programs given are “deliberately unsophisticated,” since “the best program is one written, or at least adapted, by the end user.”Nevertheless, the programs given are complete, and do their respective jobs well. The presentation in Cuthbert’s book is always lucid and wellorganized. The level is suited for senior or first-year graduate level. Circuit Design Using Personal Computers receives my highest recommendation; if there is a spot in a curriculum for a good course on modern techniques of networksynthesis, then this book would be well suited for it. Anyone interested in circuits and computer methods will find this book an absolute delight. Reprinted from /E€€ Circuits andSystems Mag., vol. 6, p. 18, Mar. 1984.
Building Expert Systems-Frederick Hayes-Roth, Donald A. Waterman, and Douglas B. Lenat, Eds. (Reading, MA: Addison-Wesley, 1983, 444 pp., $32.50.) Reviewed by John A. Biles,Rochester lnstitute of Technology, Rochester, N Y 14623. Building Expert Systems is an excellent text in probably the hottest area of artificial intelligence, if not in software in general. The book is the result of an expert systems workshop given by the editors and attended by teams representing “every developer of a significant knowledge engineering tool.” The goals of the workshop were to state and elucidate “the current state of knowledge” and toprovide “a comparative evaluation of the software tools and methodologies of knowledge engineering.” This book is the successful realizationof thesegoals, and it is the first legitimate textbook in knowledge engineering and expert systems. While Building Expert Systems is a synthesis of the work of some a coherent fiftycontributors,it reads consistently well andis textbook, not a mere concatenation of papers. The reader should have had exposure to programming and preferably to logic and artifical intelligence, although these last two topics are reviewed in the book. One warning, though, is in order. This book will not tell the reader how to buildan expert system; that would be impossible for any book. What itwilldo is acquaint the reader withthe important concepts, tools, and results of those who do buildexpert systems and thereby reduce the startup time for anyone interested in entering the arena. One recommendation is for the reader to have in mindan expert system that he or she would like to build. This proposed system need not be well thought out, but it will help provide a focus for the sometimes imposing quantity of information. Infact the editors provided a “default” expert system in their workshop by bringing in a “mystery consultant”with a proposal for an expert system to provide ”emergency management of inland oil spills.” This problem was attacked by the teams attending the workshop and provides the basis for comparisons among thedifferent techniques and approaches presented. In the absence of the reader’s own proposed system, the “inlandoilspill”problem can serve totie together some occasionally disparate threads. The book i s divided into five parts, each containing one or more chapters. The first part givesan overview and definition of expert systems and provides a succinct summary of relevant concepts from are contrasted to ordinary artificialintelligence. Expertsystems programs, and a history and classification of the more successful expert systemsis given. Thesummary of AI runs the gamut from searching techniques to inference and reasoning and is an excellent overview of thewiderfield of AI from the point of view of a knowledge engineer. The second part covers designing and building an expert system. Choices in architecture are discussed based on the characteristics of theknowledgebeing handled and the type of expert system desired. Attention also i s given to the process and methods of knowledge acquisition, in other words, how the knowledge of a human expert makes its way into a program. Finally, a comparison of knowledge engineering tools is made by comparing the inland oil spill expert systems developed by the eight teams at the workshop. The third part of the book deals with metaknowledge to help evaluate expert systems. One of the hallmarks of an expert system is that it not only can make a decision, but also can tell how and P R O C E E D I N G S OF T H E IEEE, V O L 73, NO. 2 . F E B R U A R Y 1985
whyit made its decision. This requires of the expertsystem a certain level ofknowledge about its own knowledgewhich is critical in determining if an expert system knows what it’s talking about. Recommendations for evaluating expertsystemsaremade based on case studies of existing systems. The fourth part of the book surveys programming languages and tools used i n knowledge engineering. The discussion starts with LISP, progresses to newer languages like KRL and OPS5, and then covers general tools and approaches used by knowledge engineers. The fifth and final part of the book gives the details of the inland oil spill management problem that served as the basis for the expert systems developed at the workshop and used in the book as examples. An appendix gives transcripts of the operation of six of these expert systems. Building ExpertSystemsrepresents the first true textbook in knowledge engineering. It is coherent and lucidly written, requires minimalprior experience inartificial intelligence, and covers its topic clearly and thoroughly. It will not give you a blueprint for building you first expert system, but it will point you in the right direction toward learning how to build it. Reprinted from / € E € Software, pp. 110-111. July 1984.
Book Alert
The following descriptions of recent books were prepared by the staff o f the Engineering Societies Library, 345 East 47th Street, New York, NY 10017. These books are available in the Library for loan or reference use.The prospective buyer should contact the listed publishers or his local technical book store.
VLsl Technology-S. M . Sze, Ed. (New‘fork:McCraw-Hill, 654 pp., bound, price not given, ISBN 0-07-062686-3.)
1983,
The book assumes that the reader has already acquired an introductory understanding of the physics and technology of semiconductor devices. It discusses all the important steps in the fabrication of VLSl circuits, from crystal growth to reliability testing. Each chapter describes one aspect of VLSl processing. The chapters begin with an overview of the topic, and subsequent sections deal with the basicscience underlyingindividual processsteps, the necessity for particular steps in achieving required parameters, and the tradeoffs in optimizingdevice performance and manufacturability. Thereare 400 technical illustrations and photos that can be used for practical process designs and analysis, 900 references, and 40 tables of the most accurate processes or device parameters.
Subspace Method5 of Pattern Recognition-Erkki Oja. (New York: Wiley, 1984, 187 pp., bound, price not given, ISBN 0-471-90311-6.) The present monograph discusses the fundamentals of subspace methods and the different approaches taken, concentrating on the learning subspace method used for automatic speech recognition and more generally for the classification of spectra. The subspace methods have been used in a diverse field of applications in pattern recognition and picture processing. Both the mathematics underlying the methods and their applications are reviewed. Among the mathematical preliminaries, statistical orthogonal expansions like the Principal Component and the Karhunen-Loeve expansions andtheir use in signal processing are discussed.The expansions areeasy to define butoftendifficultto compute in practice, and some numerical algorithms are related to the learning procedures for designing optimal classsubspaces for multiclass pattern recognition. Comparisons of the results obtained by the subspace classifiers andthe performances of somestandardclassifiersaremade in spectral vector classification, including extensive tests on phonemic data from word recognition with athousand word vocabulary. They indicatethatthe subspace approach to pattern recognition, for some general data types at least, is a tempting alternative. 383