Tamas Geszti, "Physical Models of Neural Networks"
1990 | ISBN-10: 9810200129 | 250 pages | PDF | 5 MB
1990 | ISBN-10: 9810200129 | 250 pages | PDF | 5 MB
PREFACE
You all know what your heart is. It is a pump. It also looks like a pump, with its muscles capable to make it expand and contract and its valves letting the
blood in and out when necessary. It is good mechanical engineering, well done for the purpose.
The same is true about your stomach: it is a chemical reactor, correctly devised and implemented for that .
But what would you say about your brain? You learn at school that it is the organ of thinking and regulation, but what should such an organ look like?
Although many other metaphores have been mentioned in this context, probably the piece of engineering closest in function to a brain is a computer. However, nothing less similar in appearance than a human brain and a man-made computer; nothing like the straightforward connection between shape and function familiar from the examples of heart and stomach.
One of the less transparent metaphores says that in some respects - at least as far as associative memory storage and recall is concerned, but probably quite a bit beyond that - the brain is analogous to a spin-glass. This suggestion was advanced by the solid-state physicist John Hopfield in 1982, and marked the beginning of an important development in creating neural network models which are tractable by tools of theoretical physics, in particular, statistical physics. The results seem to have relevance to both brain research and the so-called neural computers, providing building principles for a class of the latter.
Neural network modelling is an enormous field of which physicists' models cover only a little but fascinating corner. Those interested in the lively develop
ment of the whole area should consult recent volumes of the journal "Biological Cybernetics" and the recently launched journal "Neural Networks".
The intention of this lecture notes volume is to offer the Reader an introduc tion to the subject from the physicist's viewpoint. Although the main subject is
the Hopfield model and its descendants, feed-forward networks and their applications to computing tasks are also reviewed. The text is based on the author's course given for physics and biophysics students at Eotvos University, Budapest, in 1987 and 1988…