Reviews for Neural Networks for Pattern Recognition

1. A good introduction, lacks detail and generality.

   Rated 3 out of 5 stars, June 12th, 2004

   This is good and quite clear introduction to the field that tries to give the reader an intuitive overview to the neural networks and pattern recognition in general.

   This is a good book if you are interested in a conversationalist overview to neural networks. There are sufficient formulas to implement the algorithms, so it is good as a list of commonly used neural architectures and how they work, in a single easy-to-access place.

   However, the book is quite short and hurriedly goes through many different techniques and algorithms, giving you a brief snapshot of each one. Nice pictures abound and explanations, but the understanding that one may obtained from this book will be only superficial. Since the book does not discuss the foundations behind each technique, most of them appear disjoint and unrelated.

   Actually, the lack of detail and mathematical rigour can be confusing. The need to explain concepts intuitively is hardly an excuse, since there exist other books that manage to achieve clarity, easy of understanding and mathematical rigour, while they develop concepts with sufficient generality for the student to fully grasp the relation between various methods.

   From my own viewpoint, supervised neural network learning is just a special case of optimisation (the quantity to be optimised is the neural network parameter) under statistical uncertainty (the cost function to be minimised is only partially defined by a set of data and needs to be estimated).
   Thus, in addition to this book I also recommend taking a look at Bertseka's "Constrained optimization and Lagrange multiplier methods" and his newer "Nonlinear Pogramming" book. His "Neuro-Dynamic programming" book covers a lot more than just neural networks for pattern recognition. Advanced readers that are also interested in optimal stochastic control and reinforcement learning will find it useful.

   All in all, recommended for people that simply want to implement some neural network algorithms or for people that want a quick introduction. It is advisable, however, to keep a couple of books on estimation theory and on optimisation theory as an aid to deeper understanding.

2. Without doubt the best book available on Neural Computing.

   Rated 5 out of 5 stars, August 12th, 1999

   Bishop's book is the current bible on Neural Computing. It is superbly written and presented, and the subject material carefully selected. The ideas of neural computing are motivated from a statistical pattern recognition point of view, though the reader is not expected to have a strong foundation in probability theory - just a basic appreciation is enough to begin with. The book has enormous (though not excessive) breadth, and covers practically every aspect of tradiational neural networks, from theoretical aspects motivated by probability theory, to practical concerns about optimisation and learning, and finally to a more advanced treatement on Bayesian methods. Above all, Bishop's writing is lucid and clear, and although some of the topics are conceptually intricate, they are always readable and accessible. Buy this book if you have anything to do with neural networks!