The authors believe that SVMs are a topic now sufficiently mature that it should be viewed as its own subfield of machine learning. SVMs, first introduced by Vladimir Vapnik, are a type of linear learning machines much like the famous perceptron algorithm and, thus, function to classify input patterns by first being trained on labeled data sets (supervised learning). However, SVMs represent a significant enhancement in function over perceptrons. The power of SVMs lies in their use of nonlinear kernel functions that implicitly map input into high-dimensional feature spaces. In the high-dimensional feature spaces, linear classifications are possible; they become nonlinear in the transformation back to the original input space.

A learning problem is referred to as classification if its output take discrete values in a set of possible categories and regression if it has continuous real-valued output. A simple and useful model of an input-output functional relationship is to assume that the output variable can be expressed approximately as a linear combination of its input vector components. These linear models include the linear least squares method for regression and the logistic regression method for classification. Because a linear model has limited prediction power by itself, there has been extensive research in nonlinear models such as neural networks. However, there are two major problems with the use of nonlinear models: First, they are theoretically difficult to analyze, and second, they are computationally difficult to solve.

There are however, a variety of different approaches that claim to capture the true nature of this concept. One reason for this diversity lies in the fact that abductive reasoning occurs in a multitude of contexts. It concerns cases that cover the simplest selection of already existing hypotheses to the generation of new concepts in science. It also concerns cases where the observation is puzzling because it is novel versus cases in which the surprise concerns an anomalous observation. For example, if we wake up, and the lawn is wet, we might explain this observation by assuming that it must have rained or that the sprinklers have been on.

Several factors have led to the increase in interest in this field, which is heavily influenced by techniques from speech processing. One major factor is the recent availability of large online text collections. Another is a disillusionment with traditional AIbased approaches to parsing and natural language processing (NLP). Charniak is recognized as a distinguished contributor to what he calls traditional AI NLP, which is why it is all the more significant that in the Preface, when speaking of his recent transition to the statistical approach, he writes … few, if any, consider the traditional study of language from an artificial-intelligence point of view a "hot" area of research. A great deal of work is still done on specific NLP problems, from grammatical issues to stylistic considerations, but for me at least it is increasingly hard to believe that it will shed light on broader problems, since it has steadfastly refused to do so in the past.

The main contribution of the book Rules of Encounter: Designing Conventions for Automated Negotiation, by Jeffrey S. Rosenschein and Gilad Zlotkin, is the formulation of a principled framework within which to study interactions among artificial heterogeneous agents. This framework is based on the theory of games, which is aimed at decision problems faced by agents in situations in which the agent's welfare depends not only on its own actions but also on the actions of other agents. The examples are numerous: The personal digital assistants (PDAs) that might one day keep track of their users' itinerary will have to negotiate with PDAs of other people to adjust and synchronize their meeting schedules. Software agents looking for the right kinds of information on the Internet on behalf of their users might have to negotiate with other such agents over the access to resources. Computer agents that control a telecommunications network will have to interact with computers that control other networks and might find it beneficial to come to agreement with them.

Qualitative Reasoning: Modeling and Simulation with Incomplete Knowledge describes 15 years of research in the qualitative physics field of AI by the author and his collaborators. Qualitative physics seeks to automate human reasoning about the physical world. The original focus was on the commonsense reasoning that underlies everyday life, such as cooking with stoves, pouring coffee, parking cars, crossing streets, and playing ball. Recent work focuses on expert reasoning about scientific and engineering domains, including circuits, thermodynamics, power plants, chemical plants, and botany. Qualitative physics hypothesizes that commonsense reasoning and expert reasoning are similar enough to justify a unified treatment.

Once the topic has become well enough understood that it can be explained easily to paying customers, and stable enough that anyone teaching it is not likely to have to update his/her teaching materials every few months as new developments are reported, it can be considered to have arrived. Another reasonable indicator of the maturity of a subject, a milestone along the road to academic respectability, is the publication of a really good book on the subject--not another research monograph but a book that consolidates what is already known, surveys and relates existing ideas, and maybe even unifies some of them. Grigoris Antoniou's Nonmonotonic Reasoning is just such a milestone--well written, informative, and a good source of information on an important and complex subject. Neither is it surprising nor unreasonable that he devotes a lot of space to Reiter's (1980) default logic, which, along with Mc-Carthy's (1980) circumscription and Moore's (1985) autoepistemic logic, is one of the holy trinity of nonmonotonic reasoning. AI Magazine Volume 20 Number 3 (1999) ( AAAI) and it has been the basis of a number of different variants, all with their own strengths and weaknesses.

R B. Abhyankar Emphasizing theory and implementation issues more than specific applications and Prolog programming techniques, Computing with Logic Logic Programming with Prolog (The Benjamin Cummings Publishing Company, Menlo Park, Calif., 1988, 535 pp., $27 95) by David Maier and David S. Warren, respected researchers in logic programming, is a superb book Offering an in-depth treatment of advanced topics, the book also includes the necessary background material on logic and automatic theorem proving, making it self-contained. The only real prerequisite is a first course in data structures, although it would be helpful if the reader has also had a first course in program translation. The book has a wealth of exercises and would make an excellent textbook for advanced undergraduate or graduate students in computer science; it is also appropriate for programmers interested in the implementation of Prolog The book presents the concepts of logic programming using theory presentation, implementation, and application of Proplog, Datalog, and Prolog, three logic programming languages of increasing complexity that are based on horn clause subsets of propositional, predicate, and functional logic, respectively This incremental approach, unique to this book, is effective in conveying a thorough understanding of the subject The book consists of 12 chapters grouped into three parts (Part 1 chapters 1 to 3, Part 2. chapters 4 to 6, and Part 3 chapters 7 to 12), an appendix, and an index The three parts, each dealing with one of these logic programming languages, are organized the same First, the authors informally present the language using examples; an interpreter is also presented. Then the formal syntax and semantics for the language and logic are presented, along with soundness and completeness results for the logic and the effects of various search strategies Next, they give optimization techniques for the interpreter Each chapter ends with exercises, brief comments regarding the material in the chapter, and a bibliography Chapter I presents top-down and bottom-up interpreters for Proplog Chapter 2 offers a good discussion of the related notions: negation as failure, closed-world assumption, minimal models, and stratified programs Chapter 3 considers clause indexing and lazy concatenation as optimization techniques for the Proplog interpreter in chapter 1 Chapter 4 explains the connection between Datalog and relational algebra. Chapter 5 contains a proof of Herbrand's theorem for predicate logic.

The economic viability of a manufacturing organization depends on its ability to maximize customer services; maintain efficient, low-cost operations; and minimize total investment. These objectives conflict with one another and, thus, are difficult to achieve on an operational basis. Much of the work in the area of automated scheduling systems recognizes this problem but does not address it effectively. The work presented by this Ph.D. dissertation was motivated by the desire to generate good, costeffective schedules in dynamic and stochastic manufacturing environments (Berry 1991). Experimental analysis is used to illustrate…the PCP approach within an advanced scheduling architecture.

As I was going through them, I thought, "So many books, so few pages." AI Magazine is not a publication exclusively devoted to books, such as the New York Review or the weekly book review supplements of major newspapers. At best, it can devote a few pages each issue to book reviews. Also, it doesn't appear that frequently, just four issues a year. Given these constraints, how can the magazine best serve its readership?