American Scientist

Animal Learning and Cognition: A Neural Network Approach. Nestor A. Schmajuk. 340 pp. Cambridge University Press, 1997. $44.95.

Questions about the fundamental nature of animal learning and cognition often whet the appetite of the mathematical theorist. The reason, in part, is that psychologists who study learning in animals typically employ well-controlled paradigms, such as classical and operant conditioning, that produce tractable empirical phenomena.

In this new effort, Nestor Schmajuk turns the powerful tools of neural-network modeling to animal learning and cognition. Alhough Schmajuk is not the first or only psychologist to offer network accounts of animal learning, he is among the best and most influential modelers working in the field. He adopts E. N. Sokolov's position that animals construct neural models of environmental events expressly to predict. Discrepancies between the internal predictions generated and the events that actually transpire drive the behavior of the network and any subsequent learning. One important characteristic of Schmajuk's networks is that they can be compared to behavior in real time. He is careful to consider alternative accounts, however, and he provides the reader with ample references, along with well-crafted comparative tables.

This book fully qualifies as a primer for neural-network models of animal cognition, but it does help if you have some background or general expertise in mathematical descriptions. The writing can be terse, and the book is not recommended for the mathematically uninitiated. As a bonus, computer software accompanies the book and allows the reader to explore one of Schmajuk's models in considerable detail.—James S. Nairne, Psychological Sciences, Purdue University