Neural networks for pattern recognition /

Nigrin writes about his important fundamental research in neural networks, as treated in his dissertation and later extended. Using the seminal work of Stephen Grossberg and others on Adaptive Resonance Theory (ART) as his starting point, Nigrin addresses the question: What are the essential properties a neural network should have? He identifies 13 such properties, including the ability to self-organize using unsupervised learning, to operate in real time and under presence of noise (in the information sense), and to scale up to large problems. To achieve these, he proposes some important changes in the architecture of the networks that are currently most widely used. He then develops a model for a Self-Organizing Neural Network (SONNET) embodying his proposals. The book is extremely well written: at each stage, Nigrin takes pains to explain clearly his aims, his achievements, and improvements in future versions of SONNET. Highly recommended. Graduate through professional. R. Bharath; Northern Michigan University