The Integrated Reasoning Information Systems (IRIS) Group is a forum for international cooperation in research, development, and applications of intelligent systems. This book deals with the IRIS work on fast learning (improving the training speed of neural networks) and on invariant object recognition.

Part 1 deals with learning systems. The authors describe new algorithms and speed-up modifications to backpropagation. These advances are based on holographic networks (which use neurons that input and output a vector of complex numbers), adaptive decoupled momentum (a variant of momentum that is used to avoid small local minima during the search for solutions), detecting potentially useful relationships as new features to speed convergence, and second-order gradient methods that incorporate gradient information to change the search step length. Part 1 also deals with systolic implementations of neural networks.

Part 2 deals with object recognition systems that are invariant with respect to position, scale, and rotation effects. Topics include translation-invariant object recognition systems; networks invariant in relation to position, scale, and rotation; visual tracking; partial shape recognition using object “landmarks”; and segmentation of three-dimensional ultrasonic images.

Applications described in the book include determination of the orientation of an integrated circuit chip, recognition of aircraft silhouettes, recognition of occluded two-dimensional shapes, and analysis of three-dimensional ultrasonic images. Although the book is subtitled the sixth-generation breakthrough, the applications described are hardly awe-inspiring.

The chapters are written by different contributors, so the volume has more of the flavor of a proceedings than of a book. Chapters written by foreign authors are in somewhat stilted English. The book would be useful to an expert in neural networks who wants to learn about research being carried out by the international community. The exposition is at a research level, with little or no tutorial material.