MICHAEL R. W. DAWSON Connectionism: A Hands-On Approach Malden, MA: Blackwell Publishing, 2005, 208 pages (ISBN 1-4051-3074-1, US$69.95 Hardcover) Reviewed by JOSEPH MACINNES This book is intended to be a practical introduction to connectionist models and neural networks. The book uses as examples, simple connectionist models that have been studied over the years, with each chapter dedicated to a slightly more complex model or problem than the previous. The author also provides links to a web page where sample programs can be downloaded, allowing the reader to use these programs as they progress through the book. Although the book is well written, it falls short of its objective of being an introduction to connectionism. The book does not claim to be focused on connectionist theory, but its lack of a theoretical basis makes it difficult to follow. Throughout the book, there are references to another work by the author entitled Minds and Machines, which the reader can reference for the theoretical basis and background of connectionist models, and this may indicate the preferred use of the current book. Although this review does not include the Mind and Machines book (and this reviewer has not even read it), it is likely that Connectionism fits better as a workbook for its counterpart than it does as a standalone text. The first chapter exemplifies this problem as there is no introduction of note, and a novice reader (for whom this book is written) will be left without a clear definition of connectionism and how it relates to cognitive psychology. Much of this chapter is of the author's own history and progression from parallel distributed processing models. This chapter reads more as a prelude than introduction and would benefit from an additional chapter introducing terminology and use of connectionist models in psychology. Readers of this book should have been introduced to the major concepts before starting this text, and then directed to begin at Chapter 2. Later chapters seem to be arranged as a history of connectionism, with progressive chapters tackling later (and more advanced) concepts and implementations. Early chapters cover simple distributed memory before covering the perceptron and eventually hidden unit training. Again, the lack of a theoretical basis for connectionism makes the chapters seem as a sequential set of articles but taken on an individual basis, the chapters provide a very good layperson's description of their respective topics. The chapter on perceptrons, in particular, is an excellent introduction into what an activation function is, possible variants of activation functions, and learning rules. One of the biggest claims of the book is the inclusion of sample software, which can be downloaded from the author's website, allowing the readers to gain first-hand knowledge of three different types of networks: distributed associative memory with the James program, perceptron learning with the Rosenblatt program, and multilayer nets with the Rumelhart program. Although it is near impossible to use this book without these programs, they are not included on a CD, but the user must download them from the web, where the location and even longevity of the availability is in question. If the website ever became unavailable, the usefulness of the book would diminish drastically. …