aighb@castle.ed.ac.uk (Geoffrey Ballinger) (11/06/90)
What do people recommend in the way of a good reference book on neural networks. I am not looking for a tutorial but the maximum information in the minimum pages, preferably with formal justifications for the methods used. Mail me your replies and I will summerise to the net if there is sufficient interest, Geoff. -- Geoff Ballinger, JANET: Geoff@Uk.Ac.Ed Department of Artificial Intelligence, UUCP: ...!mcsun!ukc!Ed.Ac.Uk!Geoff Edinburgh University.
aighb@castle.ed.ac.uk (Geoffrey Ballinger) (11/11/90)
I asked ... > What do people recommend in the way of a good reference book on > neural networks. I am not looking for a tutorial but the maximum > information in the minimum pages, preferably with formal justifications > for the methods used. Mail me your replies and I will summerise to the > net if there is sufficient interest, Here are the responces I recieved ... ------------------------------------------------------------------------ > From: fraser@gov.anl.mcs.antares From a historical viewpoint, one should start with Parallel Distributed Processing, by Rumelhart and McClelland. This is a two volume set + a third that includes software. I believe that anyone who is serious about delving into neural networks should at least give these books a good looking over. If you find that these don't carry you back far enough, then there's always the so called begining in Perceptrons, by Marvin Minsky. eric. ------------------------------------------------------------------------ > From: Hakon Styri <styri@uk.ac.hw.cs> May I recommend: Simpson, Patrick K.: "Artificial Neural Systems: Foundations, Paradigms, Applications, and Implementations", Pergamon Press, 1990. It's 210 pages and gives an overview of 28 paradigms. The 'foundations' part may not be the best, but the references and bibliography makes up for that. And, the price is very nice - less than 11 pounds. ------------------------------------------------------------------------ From: Ken-ichi Maeda <tosh-km@uk.ac.ed.aiai> The following book is not a text book but a good collection of pointers. "THE 1989 NEURO-COMPUTING BIBLIOGRAPHY" (second edition), Klimasauskas, (editor), MIT Press, 1989. -- Ken ------------------------------------------------------------------------ > From: John Reynolds <reynolds@edu.bu> A collection of 43 of the most important papers, including works by William James (1890), McCulloch and Pitts (1943), Hebb (1949), von Neumann (1958), Minsky & Papert (1969), von der Malsburg (1973), Grossberg (1976, 1980), Marr and Poggio (1976, 1982), Anderson (1977), and Carver Mead (1987), among others: _Neurocomputing_ Edited by Anderson and Rosenfield: MIT Press 1988. ------------------------------------------------------------------------ > From: John Reynolds <reynolds@edu.bu> 1) A complete self-organizing neural network architecture which generates accurate eye and arm movements in response to visual signals: _Neural Dynamics of Adaptive Sensory-Motor Control_ By Stephen Grossberg and Michael Kuperstein, Pergamon Press, 1989. 2) 12 papers by leaders in the field covering questions of visual processing, pattern recognition, attentional focus, behavioral conditioning, and motor control: _Neural Networks and Natural Intelligence_ Edited by Stephen Grossberg: MIT Press, 1988. 3) A collection of 13 of Grossberg's most important papers: _Studies of Mind and Brain: Neural Principles of Learning, Perception, Development, Cognition and Motor Control_ By Stephen Grossberg: D. Reidel Publishing Company, 1982. 4) A good overview of the main models, in a standard notation, with clear and concise discussion: _Artifical Neural Systems: Foundations, Paradigms, Applications, and Implementations_ By Patrick K. Simpson, Pergamon Press, 1990. ------------------------------------------------------------------------ > From: Peter Ross <peter@uk.ac.ed.aipna> P.Wasserman, `Neural networks' - good basic book, clear about algorithms, sensible comments about pros and cons of the models it covers R. Hecht-Nielsen, `Neurocomputing' - got a lot in it, good on the maths and the groundwork, a significantly more technical read than Wasserman R.Beale+T.Jackson, `Neural Computing, an introduction' - covers less than the above two but is chatty and fairly reliable. Like any book, the odd misprint. Peter ------------------------------------------------------------------------ > From: Peter Ross <peter@uk.ac.ed.aipna> [ about ...] > Simpson, Patrick K.: > "Artificial Neural Systems: > Foundations, Paradigms, Applications, and Implementations", > Pergamon Press, > 1990. Yup, I have this too, as you say... but there are things I don't like about it as an introductory book. For example, you might not guess from the descripition of Kanerva's SDM on p.76-78 that it's closely related to the standard Hopfield net. The standardised description tends to disguise what it actually does; it has some real addressable locations where data gets stored by incrementing a bit-position-specific counter if the bit is 1 and decrementing if it is 0. The real addresses respond to any submitted address that lies within a given Hamming radius of them. The trick is to tune the radius to be just a little below half the width of the address, since in Hamming terms almost all addresses are N/2 Hamming-distant from any given address. Read the memory by submitting an address, summing-and-thresholding the counters for all real addresses that respond. The key thought is perhaps that an address decoder can be viewed as a perceptron which fires on just one address, what if we made it fire on a set of addresses? Now why didn't he say that? He obscures an otherwise simple idea. So though I like the book and it's cheap, I wouldn't like it as my only or first source. Peter ------------------------------------------------------------------------ > From: John Pelan <jp@ie.dce> Good books on Neural Nets eh ? Haven't found a decent one yet least not as you describe. I've just ordered the following; Neural Computing - an Introduction by R Beale and T Jackson (Adam Hilger - with 40% discount if you're an IOP member) July 1990 256pp Another one that I haven't yet seen but could be promising is; An Introduction to Neural Computing by Igor Aleksander (Chapman and Hill - should be out soon now, will undoubtedly be biased towards Aleksanders own work with PLN's etc but should be good ) I have read some of the following; Advances in Neural Info. processing systems 1 - ed by D Touretzky (from the IEEE 1988 conference of same name) Nice read of 'current' work but not good for beginners. Neural Computing Architectures - by Igor Aleksander ( pretty good this for PLN's and general NN stuff) Parallel Distributed Processing - Rumelhart & McClelland (A Classic in 2 volumes - everybody refers to this as it contains all the fundamentals you'll ever need, a bit thin on *exact* details but still sufficient for the determined). There's no easy way to understanding NN's (nomatter what variety) and after reading dozens of papers on the subject nothing beats actually rolling up ones sleeves and getting down to it. Good luck - we all need it I think ! Must be out of out Neural Nets :-) ------------------------------------------------------------------------ Geoff. -- Geoff Ballinger, JANET: Geoff@Uk.Ac.Ed Department of Artificial Intelligence, UUCP: ...!mcsun!ukc!Ed.Ac.Uk!Geoff Edinburgh University.