Path: utzoo!mnetor!uunet!husc6!bloom-beacon!gatech!mcnc!decvax!ucbvax!ti-csl.csc.ti.COM!NEURON-Request From: NEURON-Request@ti-csl.csc.ti.COM (NEURON-Digest moderator Michael Gately) Newsgroups: comp.ai.neural-nets Subject: NEURON Digest - V3 #3 Message-ID: <8802030935.AA12130@ucbvax.Berkeley.EDU> Date: 2 Feb 88 19:50:22 GMT Sender: usenet@ucbvax.BERKELEY.EDU Reply-To: NEURON@ti-csl.csc.ti.com Organization: The Internet Lines: 461 NEURON Digest Tue Feb 2 13:50:22 CST 1988 Volume 3 / Issue 3 Today's Topics: Re: Commercial Neural Nets Neural Net Study Group '87 neural nets proceedings Msc. Re: Commercial Neural Nets Announcing a Connectionist/Neural Network Symposium Seminar Tech report available... Reading List Suggestions ? Fault Tolerance & Neural Networks ---------------------------------------------------------------------- Date: Fri, 22 Jan 88 08:17:46 PST From: Daniel Abramovitch Can anyone recommend a good starter text on Neural Nets? I've come from the adaptive control world and neural nets seems to be the hot buzz word in industry. Thanks in advance, Danny ------------------------------ Date: 24 Jan 88 16:54:00 GMT From: codas!novavax!hcx1!brian@bikini.cis.ufl.edu Subject: Re: Commercial Neural Nets I have heard of a neural net product called MacBrain (I don't know who makes it) which runs on a Macintosh. I would be interested in hearing from people who use this product in the "real world". ------- ------- Brian M. Leach "The lasers are in the lab Harris Computer Systems The old man is dressed in white clothes 2101 W. Cypress Creek Rd. #161 Everybody says he's mad Ft. Lauderdale, FL 33309 No one knows the things he knows brian@harris.com No one knows." Neil Young, "Sedan Delivery" ------------------------------ Date: 25 Jan 88 13:43:30 PST (Monday) Subject: Neural Net Study Group From: Rockwell.HENR801c@xerox.com I'm trying to find members of Xerox in Rochester,NY interested in joinning/forming a neural net/connectionist study group. Interested parties should reply to ROCKWELL:HENR801C:XEROX. ------------------------------ Date: 25 Jan 88 08:13:36 GMT From: Mike MacGregor Subject: '87 neural nets proceedings Does anyone have an address for obtaining the proceedings from last summer's neural nets conference ? Thanks in advance. uucp: macg@alberta Innately analog: (403)432-3978 ean: macg@pembina.alberta.cdn disclaimer: I'm saving all my opinions for my thesis. ------------------------------ Date: Thu 21 Jan 88 09:32:56-PST From: Ken Laws Subject: Msc. Kaiti Riley asked about neural models of timing and phase. I would suggest the extensive recent literature on optic flow in the computational vision conferences and journals. You may want to reduce the techniques to one spatial dimension for NN experimentation, but the mathematics of spatiotemporal derivatives should be the same. A question was asked about Hopfield networks, which reminded me of the following. Hopfield nets are often hyped as a solution to the traveling salesman problem. Even allowing for the fact that only approximate solutions are found, I am not convinced that this is a good approach. Optimal paths are always nonintersecting loops, but Hopfield networks can give solutions that cross themselves. Non-NN postprocessing can correct for this by breaking and reconnecting the crossed links, but why should this be necessary? Has anyone built planarity into the constraint functions that drive the network? If this can't be done elegantly, shouldn't we prefer Karmarkar's linear programming approach or search for some other feedback solution that is more flexible? (Beam search is an AI technique that permits multiple solutions to compete, whereas Hopfield networks can only track one trajectory at a time. Perhaps the ability to consider multiple sets of solutions is critical.) Julian Dow suggested that Neuronal and Neural systems be distinguished. Unfortunately, the meanings were reversed from those already used by the neuroscientists. I refer specifically to an article in the new Daedalus issue by Jacob T. Schwartz, which mentions Neural systems as the biological ones and Neuronal as the artificial ones. While I'm on the subject, it turns out that the Daedalus AI issue is really more of a Connectionist issue. Each paper deals with either the potential of neuronal systems or the history of the symbolic vs. holistic split in AI -- all written in philosophical rather than engineering style. I particularly enjoyed Hillis' argument that each human brain requires only a gigabit of memory -- about the size of a Connection Machine! -- and [his] discussion of the emergent properties of water. I haven't quite finished the journal yet, but I recommend it to this audience. You can get copies for $5 (plus $1 outside the U.S.) from daedalus%amcad.uucp@husc6.harvard.edu or from DAEDALUS Business Office P.O. Box 515 Canton, MA 02021 -- Ken ------------------------------ Date: 27 Jan 88 15:25:14 GMT From: Dave Hampton Subject: Re: Commercial Neural Nets Regarding Commercial Neural Net Packages: At the AAAI convention last summer, I ordered a copy of a neural networks simulator for the IBM-PC called NeuralWorks Professional, from NeuralWare, Inc. The package cost just over $100 at the time, although I believe that it's selling for about $400 now. It consists of NWorks, a Neural Network construction and simulation environment, and two demonstration packages, Networks I and II. The distribution of the programs was delayed, and I received the original disks in October. It didn't work at all. A follow-up (Version 1.01) arrived in December at no cost, and I have been able to install it and get some of the simple networks running. I haven't been able to explore the package thoroughly yet, but it seems complete and is worth considering. Contact: Casimir "Casey" Klimasauskas, President NeuralWare, Inc. 103 Buckskin Court Sewickley, PA 15143 (412) 741-5959 ------------------------------ From: @C.CS.CMU.EDU:JOSE@FLASH.BELLCORE.COM Subject: Announcing a Connectionist/Neuroscience symposium Connectionist Modeling and Brain Function: The Developing Interface February 25-26, 1988 Princeton University Lewis Thomas Auditorium This symposium explores the interface between connectionist modeling and neuroscience by bringing together pairs of collaborating speakers or researchers working on related problems. Each set of speakers will provide a sample of the kind of successful interaction characterizing the rapidly developing field of computational modeling of brain function. Thursday Friday Associative Memory and Learning Sensory Development and Plasticity 9:00 am 9:00 am Introductory Remarks Preliminaries Professor G. A. Miller Announcements 9:15 am 9:15 am Olfactory Process and Associative Role of Neural Activity in the Memory: Cellular and Modeling Development of the Central Visual Studies System: Phenomena, Possible Mechanism and a Model Professor A. Gelperin Professor Michael P. Stryker AT&T Bell Laboratories University of California, San Fran Princeton University 10:30 am 10:30 am Simple Neural Models of Towards an Organizing Principle for a Classical Conditioning Perceptual Network Dr. G. Tesauro Dr. R. Linsker, Ph.D., M.D. Center for Complex Systems Research IBM Watson Research Lab Noon-Lunch Noon-Lunch 1:30 pm 1:30 pm Brain Rhythms and Network Memories: Biological Constraints on a Dynamic I. Rhythms Drive Synaptic Change Network: Somatosensory Nervous System Professor G. Lynch Dr. T. Allard University of California, Irvine University of California, San Francisco 3:00 pm 3:00 pm Brain Rhythms and Network Memories: Computer Simulation of Representational II. Rhythms Encode Memory Plasticity in Somatosensory Cortical Hierarchies Maps Professor R. Granger Professor Lief H. Finkel University of California, Irvine Rockefeller University The Neuroscience Institute 4:30 pm General Discussion 4:30 pm General Discussion 5:30 pm Reception 5:30 pm Reception Green Hall, Langfeld Lounge Green Hall, Langfeld Lounge Organizers Sponsored by Stephen J. Hanson Bellcore & Department of Psychology Princeton U. Cognitive Science Laboratory Carl R. Olson Princeton U. Human Information Processing Group George A. Miller, Princeton U. (new page) Connectionist Modeling and Brain Function: The Developing Interface February 25-26, 1988 Princeton University Lewis Thomas Auditorium Travel Information Princeton is located in central New Jersey, approximately 50 miles southwest of New York City and 45 miles northest of Philadelphia. To reach Princeton by public transportation, one usually travels through one of these cities. We recommend the following routes: By Car >From NEW YORK - - New Jersey Turnpike to Exit #9, New Brunswick; Route 18 West (approximately 1 mile) to U.S. Route #1 South, Trenton. From PHILADELPHIA - - Interstate 95 to U.S. Route #1 North. From Washington - - New Jersey Turnpike to Exit #8, Hightstown; Route 571. Princeton University is located one mile west of U.S. Route #1. It can be reached via Washington Road, which crosses U.S. Route #1 at the Penns Neck Intersection. By Train Take Amtrak or New Jersey Transit train to Princeton Junction, from which you can ride the shuttle train (known locally as the "Dinky") into Princeton. Please consult the Campus Map below for directions on walking to Lewis Thomas Hall from the Dinky Station. For any further information concerning the conference please contact our conference planner: Ms. Shari Landes Psychology Department Princeton University, 08540 Phone: 609-452-4663 Elec. Mail: shari@mind.princeton.edu ------------------------------ Date: Thu, 28 Jan 88 08:28:44 CST From: @RELAY.CS.NET:UNICORN!LUSE@NOSC.MIL Subject: Seminar ACM SIGANS presents "Economic Prediction using Neural Networks" Dr. Halbert White Professor of Economics at UCSD Tuesday February 23 6-8 pm General Dynamics CRA Pavillion For more information call Dave Holden at 592-5026. GD CRA Pavillion is located in Missile Park, just east of 163 off Clairemont Mesa Blvd. (Thomas Bros. map page 45, F6.) ------------------------------ Date: Wed 20 Jan 88 12:15:45-CST From: Jim Anderson Subject: Tech report available... MCC-EI-287-87 Neural Networks and NP-complete Optimization Problems; A Performance Study on the Graph Bisection Problem Carsten Peterson and James R. Anderson Microelectronics and Computer Technology Corporation 3500 West Balcones Center Drive Austin, TX 78759-6509 Abstract: The performance of a mean field theory (MFT) neural network technique for finding approximate solutions to optimization problems is investigated for the case of the minimum cut graph bisection problem, which is NP-complete. We address the issues of solution quality, programming complexity, convergence times and scalability. Both standard random graphs and more structured geometric graphs are considered. We find very encouraging results for all these aspects for bisection of graphs with sizes ranging from 20 to 2000 vertices. Solution quality appears to be competitive with other methods, and the effort required to apply the MFT method is minimal. Although the MFT neural network approach is inherently a parallel method, we find that the MFT algorithm executes in less time than other approaches even when it is simulated in a serial manner. --------------------------------------------------------------------------- Requests for copies should include name and land address. Please send requests to HINER@MCC.COM. ------------------------------ Date: Tue, 26 Jan 88 11:00:35 CST From: @C.CS.CMU.EDU:CECI@BOULDER.COLORADO.EDU Subject: Re: Reading list suggestions? Here's a suggestion. It's a short work, mostly conceptual, and it concerns some of the problems coonectionist models can be expected to encounter when trying to do natural language processing. David L. Waltz. Connectionist Models: Not Just a Notational Variant, Not a Panacea. Abstract: Connectionist models inherently include features and exhibit behaviors which are difficult to achieve with traditional logic- based models. Among the more important characteristics are: (1) the ability to compute nearest match rather than requiring unification or exact match; (2) learning; (3) fault tolerance through the integration of overlapping modules, each of which may be incomplete or fallible, and (4) the possibility of scaling up such systems by many orders of magnitude, to operate more rapidly or to handle much larger problems, or both. However, it is unlikely that connectionist models will be able to learn all of language from experience, because it is unlikely that a full cognitive system could be built via learning from an initially random network; any successful large-scale connectionist learning system will have to be to some degree "genetically" prewired. The paper is only seven pages long, including references, so you can get an idea of the depth of analysis. It is, however, very clearly written and sets out the major obstacles connectionist language learning will have to overcome. Unfortunately, I don't have the complete source citation; it's a technical report, but I don't remember who put it out. Waltz listed his credentials as "Thinking Machines Corporation and Brandeis University," so perhaps one of those two institutions will know the exact citation. I'd be very interested in the reading list you compile. Cheers, Lou Ceci Dept. of Journalism and Mass Communications Univ. of Northern Colorado Greeley, CO 80639 (303) 351-2726. home: 3065 30th St. #6 Boulder, CO 80301 (303) 449-7839 ------------------------------ Date: Fri, 29 Jan 88 23:49:42 pst From: "Andrew J. Worth" Subject: Fault Tolerance & Neural Networks I would like to thank those who responded to my query and re-post this query for information on: - the inherent fault-tolerance in neural networks - determining the fault-tolerance capabilities of neural networks - increasing fault tolerance in neural networks - using neural networks for traditional fault tolerance applications Results of this query so far follow. Due to address problems, I am re-posting this request for information. Anyone with additional information on the above subjects is encouraged to respond via one of my addresses given below. Thanks in advance. ----------------------------------------------------------------------- From: kurt@bach.csg.uiuc.edu (Kurt) T. Hogg and B. Huberman, "Understanding Biological Computation: Reliable Learning and Recognition," Proceeding of the National Academy of Science, November 1984, pp. 6871-6875. ----------------------------------- From: ee.worden@a20.cc.utexas.edu (Sue J. Worden) C. R. Legendy, "On the scheme by which the human brain stores information," MATH.BIOSCI., vol. 1, pp. 555-597, 1967 J. J. Hopfield, "Neural networks and physical systems with emergent collective computational abilities," PROC>NATL. ACAD.SCI.USA, vol. 79, no. 8, pp. 2554-2558, 1982 J. A. Anderson, "Cognitive and psychological computation with neural models," IEEE TRANS.SYST.,MAN,CYBERN., vol. SMC-13, no. 5, pp. 799-815, 1983. S. S. Venkatesh, "Epsilon capacity of neural networks," NEURAL NETWORKS FOR COMPUTING, AIP CONF.PROC. 151, J. S.Denker, ed., pp. 440-445. 1986 OTHER POSSIBILITIES: ----------------------------------- see PDP ch 12 p.472 & PDP ch 7 p.303 & PDP ch22 p.413 PDP = Rumelhart, D., and McClelland, J., Parallel Distributed Processing: Explorations in the Microstructure of Cognition: Volumes 1 and 2, Bradford Books/MIT Press, Cambridge, 1986 ----------------------------------- A mention of using Hopfield nets for FT applications: correcting serial transmissions? (see Lippmann, Richard P., An Introduction to Computing with Neural Netw, IEEE ASSP, April 1987, p. 8.) ----------------------------------- Simplson references Cottrell about gracefull degridation of brains: Cottress, G. and Small, S., "Viewing Parsing as Word Sense Discrimination: A Connectionis Approach", Computational Models of Natural Language Processing. Bara, B. and Guida, G. (Eds.), Elsevier Science Publishers, B.B.: North-Holland (1984). ONGOING RESEARCH: ----------------------------------- Sue J. Worden, U. Texas, Austin. -fault tolerance of a neural network architecture based on compacta theory (see ref. above to C. R. Legendy) Michael J. Carter, U. New Hampshire. -a quantitative theory of fault tolerance for neural networks initially for multi-layer perceptrons. ----------------------------------------------------------------------- -Andy "everyone just says they are fault tolerant and that's all" worth@iris.ucdavis.edu worth%iris.ucdavis.edu@relay.cs.net worth@clover.ucdavis.edu 1421 H Street Apt 4, Davis, CA, 95616-1128 (916) 753-9910 ------------------------------ End of NEURON-Digest ********************