schneide@arisia.Xerox.COM (Kris A. Schneider) (01/05/91)
Hi folks, I'm in need of finding any literature or reference materials that deal with neural networks used by an expert system. Any help? Thanks much! -Kris schneide@arisia.xerox.com
pratt@paul.rutgers.edu (Lorien Y. Pratt) (01/07/91)
Hi,
There hasn't been a lot done to date on integrating neural networks and
expert systems. What has been done falls into a few of categories:
1) Using neural networks as knowledge elicitation devices
2) Incorporating domain knowledge into nets to improve performance
3) Using networks in traditional ``expert systems'' tasks
1) is by far the least studied. One reference is:
@misc{ fu-85,
MYKEY = " fu-85 : .unb .con .bap .csy ",
TITLE = "{Learning Intermediate Concepts in Constructing a
Hierarchical Knowledge Base}",
KEY = "fu-85",
This was in the proceedings of the IEEE IEA/AIE conference, in 1989 I
believe.
2) This is an exciting problem. It's been recently addressed by myself
(unpublished so far) and:
@techreport{ shavlik-89b,
MYKEY = " shavlik-89b : .bap .con .csy .rep .app ",
TITLE = "{Combining Explanation-Based and Neural Learning:
An algorithm and Empirical Resulst}",
AUTHOR = "J. W. Shavlik and G. G. Towell",
YEAR = 1989,
NUMBER = 859,
MONTH = "June",
INSTITUTION = "Department of Computer Science, University of
Wisconsin",
ADDRESS = "Madison, Wisconsin",
3) Lots of people use expert systems for classification tasks.
Back-propagation networks also solve classification tasks, and many data
sets have had both methods applied, and the results compared (refs available).
So many papers that don't superficially appear relevant to expert systems are
in fact, if you're interested in classification tasks. An excellent recent
overview and comparison is the recent Morgan Kauffmann book by Weiss and
Kulikowski, called ``Computer Systems that Learn''. They cover both neural
networks and expert systems, and discuss the advantages/disadvantages of both
approaches, and conclude with a page saying it would be nice to have hybrids.
Another reference is:
@article{ gallant-88,
MYKEY = " gallant-88 : .med .app .bap .csy .con ",
TITLE = "{Connectionist Expert Systems}",
author = "Stephen I. Gallant",
journal = "{Communications of the ACM}",
volume = 31,
number = 2,
pages = {152--169},
year = 1988
}
Although this is fairly preliminary work, it does also discuss the connection
you're interested in.
I hope these references are enough to get you started; there are several more
when you're finished with these.
--Lori Pratt
--
-------------------------------------------------------------------
L. Y. Pratt Computer Science Department
pratt@paul.rutgers.edu Rutgers University
Hill Center
(201) 932-4634 (Hill Center office) New Brunswick, NJ 08903, USA
(201) 846-4766 (home)guedalia@bimacs.BITNET (David Guedalia) (01/07/91)
In article <15385@arisia.Xerox.COM> schneide@arisia.Xerox.COM (Kris A. Schneider) writes: >Hi folks, > >I'm in need of finding any literature or reference materials that deal with neural networks used by an expert system. Any help? T > >-Kris >schneide@arisia.xerox.com Hi, I am sorry I do not know of any refrences. But from what I understand of Expert Systems and Neural Nets they seem to be opposites. (I would be intrested of hearing of an aplication of the above system). Expert Systems is the art of taking human information and encoding it into a computer. While Neural Networks try to devlop there own representation of the information. Any comments ? David
bryden@brahms.udel.edu (Chris Bryden) (01/08/91)
In article <> guedalia@bimacs.UUCP (David Guedalia) writes: }In article <> schneide@arisia.Xerox.COM (Kris A. Schneider) writes: }>...neural networks used by an expert system. Any help? } } Hi, I am sorry I do not know of any refrences. But from what I }understand of Expert Systems and Neural Nets they seem to be }opposites. (I would be intrested of hearing of an aplication of the }above system). Actually, I heard a nasty rumor that someone had a method by which they could pull rules, for an expert system, from a properly trained neural network. I would appreciate any references that would make or break this rumor. -- {gateway}!udel!brahms!bryden | I am a direct result of the policies and actions bryden@udel.edu 302-451-6339 | that are endorsed by the University of Delaware.
rr2p+@andrew.cmu.edu (Richard Dale Romero) (01/09/91)
Chris Bryden writes: >Actually, I heard a nasty rumor that someone had a method by which >they could pull rules, for an expert system, from a properly trained >neural network. I would appreciate any references that would make >or break this rumor. hmmm... sounds like a small, local lecture i attended for a new visiting professor here at cmu. basically, he took a few rules that described an approximation of a system, (he was using a 3-d surface map,) and created a neural network that conformed to those rules exactly. the rules were like: given x in <high> and y in <medium>, z will be in <low>; given x in <low>, z will be in <high>. he made a couple layers that acted as a classifier, then he made a couple more that applied the rules. the whole thing was done by hand, but he said that the whole thing could be automated, which doesn't seem too far-fetched. anyways, then, he trained the whole thing on the actual surface map using back propagation. because of the original rules that were encoded into the network, the new network was fairly easy to pull *more refined* rules out of. the rules were implicit in the original network architecture. basically, the original rules had used linear classifications, and it was evident, after examining the trained network, that non-linear rules had been necessary to encode the surface map. there were some problems raised with the method, though. it was evident that if your original rules were "bad" or you got them from a poor expert, that the rules that would be developed would not match the original form they were encoded with. in other words, if you introduced several erroneous rules, you may never realize that these rules had been changed into something totally different. for example, suppose you had the classification <high-medium> for several rules, but it ended up that this classification wasn't necessary for any rules. the network would learn to ignore this classification and it would be hard to realize that this was occurring. suffice it to say, that the pulling out of these more refined rules is not fool-proof. sorry i don't have a reference, but i'm not even sure that an entire paper has been completed. -rick
esrmm@warwick.ac.uk (Denis Anthony) (01/09/91)
In article <17355@brahms.udel.edu> bryden@brahms.udel.edu (Chris Bryden) writes: >In article <> guedalia@bimacs.UUCP (David Guedalia) writes: >}In article <> schneide@arisia.Xerox.COM (Kris A. Schneider) writes: >}>...neural networks used by an expert system. Any help? >} >} Hi, I am sorry I do not know of any refrences. But from what I >}understand of Expert Systems and Neural Nets they seem to be >}opposites. (I would be intrested of hearing of an aplication of the >}above system). > >Actually, I heard a nasty rumor that someone had a method by which >they could pull rules, for an expert system, from a properly trained >neural network. I would appreciate any references that would make >or break this rumor. > >-- One I have read is :- %Q Saito K, and Nakano R %D 1990 %T Rule Extraction from Facts and Neural Networks %J INNS Int. Neural Networks Conference (Paris) %V 1 %P 379-382 Denis
csa18@seq1.keele.ac.uk (R.J. Husmo) (01/09/91)
>In article <15385@arisia.Xerox.COM> schneide@arisia.Xerox.COM (Kris A. Schneider) writes: >>Hi folks, >> >>I'm in need of finding any literature or reference materials that deal with neural networks used by an expert system. Any help? T >> >>-Kris >>schneide@arisia.xerox.com There are very few papers on the subject of artificial neural networks (ANN) and experts systems. My Master's dissertation concerned the relationship between knowledge and ANNs. I didn't find any papers which were particularly useful. I had to invent methods for interfacing between ESs and ANNs. In short, there is no easy way of extracting rules from a medium/large ANN. For smaller, binary and layered, networks there are several methods, the easiest of which is described below: Present all possible input combinations to the network (from 000000000, say, to 111111111). For each output node, record all input combinations which turn this node ON. Reduce these lists. The result should be a list of input values which turn on an output node, (i.e. disjunctive normal form) eg: Oa is on IF Ia and not Ib OR Ia and Ic and Id OR Id. (Oa = output node a, Ix Input node x) There are several interesting issues here. What criteria is used to decide whether a node is on or off? What is the easiest way of reducing the humonguous lists produced by the method described above? What do we do if we have more than 12 input nodes (2^12 input combinations is enough...)? How do we handle fuzziness? etc, etc... These are, of course, only some issues revealed by my research. I may write a paper about the findings one rainy day. In article <2670@bimacs.BITNET> guedalia@bimacs.UUCP (David Guedalia) writes: > Hi, I am sorry I do not know of any refrences. But from what I >understand of Expert Systems and Neural Nets they seem to be >opposites. (I would be intrested of hearing of an aplication of the >above system). > Expert Systems is the art of taking human information and encoding >it into a computer. While Neural Networks try to devlop there own >representation of the information. > > Any comments ? This is correct. The term expert system is, however, very often used synonomously with knowledge based systems. And quite a lot of these system will, like ANNs, infer their rules from the data available. I hope the above wasn't too incoherent, but as I've changed research field now, I haven't got all that much time too spend spel-checking etc. Reidar.
sestito@latcs1.oz.au (Sabrina Sestito) (01/15/91)
In article <17355@brahms.udel.edu> bryden@brahms.udel.edu (Chris Bryden) writes: >In article <> guedalia@bimacs.UUCP (David Guedalia) writes: >}In article <> schneide@arisia.Xerox.COM (Kris A. Schneider) writes: >}>...neural networks used by an expert system. Any help? >} > ....... > >Actually, I heard a nasty rumor that someone had a method by which >they could pull rules, for an expert system, from a properly trained >neural network. I would appreciate any references that would make >or break this rumor. > There is definitely some work going on in this area. My Ph.D research topic is concerned with developing a method which will extract rules from a neural network. Some of our published papers are: "Machine learning using single-layered and multi-layered neural networks" by Sabrina Sestito and Tharam Dillon, in Tools for AI, Wahington D.C., 1990 "Using multi-layered neural networks for learning symbolic knowledge" by Sabrina Sestito and Tharam Dillon, in AI'90, Perth, Australia, 1990. The extension of this work, to determine disjunctive rules has also been developed, but at this stage has not been published. Any comments would be appreciated. Sabrina Sestito (sestito@latcs1.oz)
sci240s@monu6.cc.monash.edu.au (mr w.j. ho) (01/15/91)
sestito@latcs1.oz.au (Sabrina Sestito) writes: >In article <17355@brahms.udel.edu> bryden@brahms.udel.edu (Chris Bryden) writes: >>In article <> guedalia@bimacs.UUCP (David Guedalia) writes: >>}In article <> schneide@arisia.Xerox.COM (Kris A. Schneider) writes: >>}>...neural networks used by an expert system. Any help? >>} >> ....... >> >>Actually, I heard a nasty rumor that someone had a method by which >>they could pull rules, for an expert system, from a properly trained ^^^^^^^^^^^^^^^^This sound similar to what the package by HNC INC is suppose to do. The package is KnowledgeNet and it is a supplement to their mainstream NN packages (eg, ExploreNet 3000). I do not know whether it is available yet as the last time I heard about it was it is suppose to be released very soon. >>neural network. I would appreciate any references that would make >>or break this rumor. >> > Any comments would be appreciated. > Sabrina Sestito > (sestito@latcs1.oz) -- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^ Wey Jing Ho Tel: 61-3-5732567 E-mail : sci240s@monu6.cc.monash.edu.au ^ ^ Physics Dept., Monash University ( Caulfield Campus ), Melbourne, AUSTRALIA ^ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
rxs3707@tesla.njit.edu (01/16/91)
A few months back, I was writing a survey paper on Expert Systems
based on Neural Networks. I guess, there are very few references
on this subject at the moment. Searching through Science Citation
Index(SCI), the ultimate tool for finding references, could yield
only the following. Perhaps, you may like to through SCI again.
- Raj
PRIMARY REFERENCES:
1. Dietz W.E., Kiech E.L., Ali M. (1989) Jet and Rocket Enginer Fault
Diagnosis in Real Time. Journal of Neural Network Computing, Volume 1,
Number 1, 5-18
2. Yoon Youngohc, Brobst R.W. Bergstresser P.R., Peterson. L.L.(1989)
A desktop Neural Network for Dermatology Diagnosis. Journal of Neural Network
Computing, Volume 1, Number 1, 43-52
3. Gallant S.I. (1988) Connectionist Expert Systems, Communications of the
ACM. Volume 3, Number 2, 152-169
4. DARPA Neural Network Study, Chapter-19, Comparison with other Technologies.
(1989) 195-229
SECONDARY REFERENCES:
7. Norman D.A. (1988) Conclusions on PDP approach, Parellel Distributed
Processing. MIT press, vol-2 532-552
8. Johnson D. (1987) More approaches to Travelling Salesman Guide.
Nature, vol 330, Dec 1987
9. Denning Peter J.(1988) Blindness in designing Intelligent Systems, American
Scientist. Vol 76, 118-119
10. Piscane V.L., Silwa N. (1988) Computer Systems, Aerospace America.
Dec 1988, 68-69
\end {document}