neuron-request@HPLABS.HP.COM (Neuron-Digest Moderator Peter Marvit) (11/16/88)
Neuron Digest Tuesday, 15 Nov 1988 Volume 4 : Issue 24 Today's Topics: Re: GENETIC LEARNING ALGORITHMS neural nets and finite elements Re: Schedule of remaining neural network talks this semester Re: PDP prog:s source code fo 88 Connectionist Proceedings Response to "Learning with NNs" Frontiers in Neuroscience Object-Oriented Languages for NN Description seperability and unbalanced data discussion Cyberspace Implementation Issues (Tee into optic nerve?) CA simulator for suns: ftp up again AI Geneology Send submissions, questions, address maintenance and requests for old issues to "neuron-request@hplabs.hp.com" or "{any backbone,uunet}!hplabs!neuron-request" ------------------------------------------------------------ Subject: Re: GENETIC LEARNING ALGORITHMS From: brian@caen.engin.umich.edu (Brian Holtz) Organization: U of M Engineering, Ann Arbor, Mich. Date: 02 Nov 88 23:48:00 +0000 Does anyone know of any references that describe classifier systems whose messages are composed of digits that may take more than two values? For instance, I want to use a genetic algorithm to train a classifier system to induce lexical gender rules in Latin. Has any work been done on managing the complexity of going beyond binary-coded messages, or (better yet) encoding characters in messages in a useful, non-ASCIIish way? I will summarize and post any responses. In article <7104@bloom-beacon.MIT.EDU>, thefool@athena.mit.edu (Michael A. de la Maza) writes: > > Has anyone compiled a bibliography of gla articles/books? In "Classifier Systems and Genetic Algorithms" (Cognitive Science and Machine Intelligence Laboratory Technical Report No. 8) Holland lists some 80 or so applications of GAs, and offers a complete bibliography to interested parties. He can be reached at the EECS Dept., Univ. of Michigan, Ann Arbor MI 48109 (he doesn't seem to have an obvious email address here...). You can get a copy of the technical report from Sharon_Doyle@ub.cc.umich.edu. ------------------------------ Subject: neural nets and finite elements From: buc@Jessica.stanford.edu (Robert Richards) Organization: Stanford University Date: 05 Nov 88 21:28:12 +0000 Does anyone have any information or pointers to articles which deal with the use of neural net in finite element analysis? I am especially interested in the application of genetic algorithms, specifically simulated annealing, to determine the optimal geometry of the part being modeled. Thank you in advance. Rob Richards Stanford University ------------------------------ Subject: Re: Schedule of remaining neural network talks this semester From: tek@cmx.npac.syr.edu (Aman U. Joshi) Organization: Northeast Parallel Architectures Center, Syracuse NY Date: 06 Nov 88 23:43:33 +0000 In article <1072@cseg.uucp> are@hcx.uucp (ALAN RAY ENGLAND) writes: > > >On 11/18/88 E. Tzanakou will give a talk titled "ALOPEX: Another >optimization method." As a PhD student studying optimal learning in >neural networks I am intrigued by the title. Could someone enlighten me >as to exactly what ALOPEX is. A reference to a publication would be >greatly appreciated. > I am working with Prof. Erich Harth, the inventor of "ALOPEX", at Syracuse University, Syracuse, NY. I have used his technique for 3-d crystal formation, for pattern recognition, and (presently) for VLSI Standard Cell Placement. "ALOPEX" originated as an abbreviation of "ALgorithm fOr Pattern EXtraction", but also means 'fox' in greek. I have had the opportunity to run this on (perhaps the world's fastest) machine, The Connection Machine-2, by Thinking Machine Inc. The results are fantastic. Those interested in "ALOPEX"(an inherently parallel stochastic optimization technique) should contact Prof. Erich Harth, 201, Physics Building,Syracuse University, Syracuse, NY -13244. or, e-mail me at tek@cmx.npac.syr.edu. Prof. Harth's phone number :(315) 443-2565. Thank you, Aman U. Joshi, (315) 443-3573. aujoshi@sunlab.npac.syr.edu ------------------------------ Subject: Re: PDP prog:s source code fo From: spl@cup.portal.com (Shawn P Legrand) Organization: The Portal System (TM) Date: 08 Nov 88 21:59:49 +0000 For those looking for source code to Rumelhart's et. al. PDP volumes check out the "3rd volume" - "Handbook for PDP". This book (spiral bound paperback) comes with two diskettes of C source code dealing with topics covered in the original two volumes. If you would like more specific information on this volume send me some EMail and I would be happy to respond. Shawn P. Legrand, CCP +----------------------------+ | spl@cup.portal.com | | or | | ...sun!cup.portal.com!spl | +----------------------------+ [[ Editor's Note: Is it available via ftp somewhere? I also haven't seen the availability of a MAC version. Could someone enlighten the Digest? -PM ]] ------------------------------ Subject: 88 Connectionist Proceedings From: terry@cs.jhu.edu (Terry Sejnowski <terry@cs.jhu.edu>) Date: Tue, 08 Nov 88 19:41:15 -0500 NOW AVAILABLE: Proceedings of the 1988 Connectionist Models Summer School, edited by David Touretzky, Geoffrey Hinton, and Terrence Sejnowski. Available from: Morgan Kaufmann Publishers, Inc. Order Fulfillment Center P.O. Box 50490 Palo Alto, CA 94303-9953 tel. 415-965-4081 Cost is $24.95 plus $2.25 postage and handling ($4.00 for foreign orders.) For each additional volume ordered, increase postage by $1.00 (foreign, $3.00). Enclose full payment by check or money order. California residents please add sales tax. Terry ------------------------------ Subject: Response to "Learning with NNs" From: goodhart@cod.nosc.mil (Curtis L. Goodhart) Date: Wed, 09 Nov 88 08:01:45 -0800 >Subject: Learning with NNs >From: Dario Ringach >dario%TECHUNIX.BITNET@CUNYVM.CUNY.EDU> >Date: Wed, 19 Oct 88 13:36:32 +0200 > >Has anyone tried to approach the problem of learning in NNs from a >computability-theory point of view? For instance, let's suppose we use a >multilayer perceptron for classification purposes. What is the class of >discrimination functions learnable with a polynomial number of examples >such that the probability of misclassification will be less than P (using a >determined learning algorithm, such as back-prop)? > >It seems to me that these type of questions are of importance if we really >want to compare between different learning algorithms, and computational >models. > >Does anyone have references to such a work? Any references will be >appreciated! One reference addressing computability in neural networks is the section on "Computable Functions and Complexity in Neural Networks" by Omer Egecioglu, Terrence R. Smith and John Moody in the book "Real Brains Artificial Minds" by John L. Casti, and Aders Kalquist (1987). Publisher is North-Holland. Curtis L. Goodhart ------------------------------ Subject: Frontiers in Neuroscience From: terry@cs.jhu.edu (Terry Sejnowski <terry@cs.jhu.edu>) Date: Wed, 09 Nov 88 19:06:29 -0500 The latest issue of Science (4 November) has a special section on Frontiers in Neuroscience. The cover is a spectacular image of a Purkinje cell by Dave Tank. Four of the major reviews in the issue make contact with network modeling: Tom Brown et al. on Long-Term Synatpic Potentiation; Steve Lisberger on The Neural Basis for Learning of Simple Motor Skills; Steve Wise and Bob Desimone on Insights into Seeing and Grasping; and Pat Churchland and Terry Sejnowski on Perspectives on Cognitive Neuroscience. See also the letter by Dave Tank et al. on Spatially Resolved Calcium Dynamics of Mammalian Purkinje Cells in Cerebellar Slice. This issue was timed to coincide with the Annual Meeting of the Society for Neuroscience in Toronto next week. Terry ------------------------------ Subject: Object-Oriented Languages for NN Description From: Dario Ringach <dario%TECHUNIX.BITNET@CUNYVM.CUNY.EDU> Date: Fri, 11 Nov 88 09:58:15 +0200 Can anyone provide me references to high-level description languages (preferably Object-Oriented ones, like the P3 simulation system) for the description of NN models? What about CST (Concurrent SmallTalk)? It seems ideal for the description of NNs to be simulated in massive parallel fine-grained architectures such as the J-Machine ... Has anyone any experience to share on this topic? Thanks in advance. - --Dario. ------------------------------ Subject: seperability and unbalanced data discussion From: Richard Rohwer <rr%eusip.edinburgh.ac.uk@NSS.Cs.Ucl.AC.UK> Date: Fri, 11 Nov 88 11:29:38 +0000 In a close inspection of convergence ailments afflicting a multilayer net, I found that the problem boiled down to a layer which needed to learn the separable AND function, but wasn't. So I had a close look at the LMS error function for AND, in terms of the the weights from each of the two inputs, the bias weight, and the multiplicities of each of the 4 exemplars in the truth table. It turns out that the error can not be made exactly 0 (with finite weights), so minimization of the error involves a tradeoff between the contributions of the 4 exemplars, and this tradeoff is strongly influenced by the multiplicities. It is not difficult to find the minimum analytically in this problem, so I was able to verify that with my highly unbalanced training data, the actual minimum was precisely where the LMS algorithm had terminated, miles away from a reasonable solution for AND. I also found that balanced data puts the minimum where it "belongs". The relative importance of the different exemplars in the LMS error function runs as the square root of the ratio of their multiplicities. So I solved my particular problem by turning to a quartic error function, for which it is the 4th root of this ratio that matters. (The p-norm, p-th root of the sum of the p-th powers, approaches the MAX norm as p approaches infinity, and 4 is much closer to infinity than 2.) ---Richard Rohwer, CSTR, Edinburgh ------------------------------ Subject: Cyberspace Implementation Issues (Tee into optic nerve?) From: "RCSDY::YOUNG%gmr.com"@RELAY.CS.NET Date: Fri, 11 Nov 88 09:22:00 -0400 In article <10044@srcsip.UUCP> lowry@srcsip.UUCP () writes: > If you could "tee" into the optic nerve, it seems like you could feed > in pre-digested data at a much lower rate. In Neuron Digest V4 #19, jdb9608@ultb.UUCP asks in response: > I'm cross-posting ... to the neural-nets group in the hope > that someone there will have some comment or idea on how a computer could > possibly generate a consensual hallucination for its operator, hopefully > entirely within the operator's mind. In my thesis work [Young, R. A. , Some observations on temporal coding of color vision: psychophysical results,Vision Res. 17, 957-965 (1977)] I electrically stimulated the eyes of four volunteer subjects and got phosphene colors that were specific to the electrical pulse stimulus patterns. In that paper I reference a number of other papers in the psychology and psychophysical literature regarding the production of patterns and colors via direct electrical stimulation of the eye. ------------------------------ Subject: CA simulator for suns: ftp up again From: cgl%raven@lanl.gov (Chris Langton ) Date: Fri, 11 Nov 88 10:06:14 -0700 The CA simulator for Suns is available again via ftp in the post-virus world. My earlier message gave an example of how to obtain it via anonymous ftp from 128.165.96.120. The tar file is in the "pub" directory entitled "cellsim_1.0.tar". If you are unsure about anonymous ftp, send me a message and I will send you a sample script. Again, Europeans should obtain a copy from Tom Buckley at Leeds University: Dr. T.F. Buckley Department of Computer Studies Leeds University Leeds, LS2 9JT England, UK email: buckley%uk.ac.leeds.dcs@UKACRL Chris Langton Center for Nonlinear Studies Phone: 505-665-0059 MS B258 Email: cgl@LANL.GOV Los Alamos National Laboratory Los Alamos, New Mexico 87545 [[ Editor's Note: Several people asked me about the Cellular Automata mailing list. Send mail to CA-request@think.com to be added to the list. It is relatively low volume nowadays, but I'm sure Neuron readers could spice things up ;-) -PM ]] ------------------------------ Date: Thu, 3 Nov 88 09:35:44 PST From: rik%cs@ucsd.edu (Rik Belew) Subject: AI family tree AI GENEALOGY Building an AI family tree Over the past several years we have been developing a collection of bibliographic references to the literature of artificial intelligence and cognitive science. We are also in the process of developing a system, called BIBLIO, to make this information available to researchers over Internet. My initial work was aimed at developing INDEXING methods which would allow access to these citations by appropriate keywords. More recently, we have explored the use of inter-document CITATIONS, made by the author of one document to previous articles, and TAXONOMIC CLASSIFICATIONS, developed by editors and librarians to describe the entire literature. We would now like to augment this database of bibliographic information with "cultural" information, specifically a family tree of the intellectual lineage of the authors. I propose to operationalize this tree in terms of each author's THESIS ADVISOR and COMMITTEE MEMBERS, and also the RESEARCH INSTITUTIONS where they work. It is our thesis that this factual information, in conjuction with bibliographic information about the AI literature, can be used to characterize important intellectual developments within AI, and thereby provide evidence about general processes of scientific discovery. A nice practical consequence is that it will help to make information retrievals from bibliographic databases, using BIBLIO, smarter. I am sending a query out to several EMail lists to ask for your help in this enterprise. If you have a Ph.D. and consider yourself a researcher in AI, I would like you to send me information about where you got your degree, who your advisor and committee members were, and where you have worked since then. Also, please forward this query to any of your colleagues that may not see this mailing list. The specific questions are contained in a brief questionnaire below, and this is followed by an example. I would appreciate it if you could "snip" this (soft copy) questionnaire, fill it in and send back to me intact because this will make my parsing job easier. Also, if you know some of these facts about your advisor (committee members), and their advisors, etc., I would appreciate it if you could send me that information as well. One of my goals is to trace the genealogy of today's researchers back as far as possible, to (for example) participants in the Dartmouth conference of 1956, as well as connections to other disciplines. If you do have any of this information, simply duplicate the questionnaire and fill in a separate copy for each person. Let me anticipate some concerns you may have. First, I apologize for the Ph.D. bias. It is most certainly not meant to suggest that only Ph.D.'s are involved in AI research. Rather, it is a simplification designed to make the notion of "lineage" more precise. Also, be advised that this is very much a not-for-profit operation. The results of this query will be combined (into an "AI family tree") and made publically available as part of our BIBLIO system. If you have any questions, or suggestions, please let me know. Thank you for your help. Richard K. Belew Asst. Professor Computer Science & Engr. Dept. (C-014) Univ. Calif. - San Diego La Jolla, CA 92093 619/534-2601 619/534-5948 (messages) rik%cs@ucsd.edu -------------------------------------------------------------- AI Genealogy questionnaire Please complete and return to: rik%cs@ucsd.edu NAME: Ph.D. year: Ph.D. thesis title: Department: University: Univ. location: Thesis advisor: Advisor's department: Committee member: Member's department: Committee member: Member's department: Committee member: Member's department: Committee member: Member's department: Committee member: Member's department: Committee member: Member's department: Research institution: Inst. location: Dates: Research institution: Inst. location: Dates: Research institution: Inst. location: Dates: -------------------------------------------------------------- AI Genealogy questionnaire EXAMPLE NAME: Richard K. Belew Ph.D. year: 1986 Ph.D. thesis title: Adaptive information retrieval: machine learning in associative networks Department: Computer & Communication Sciences (CCS) University: University of Michigan Univ. location: Ann Arbor, Michigan Thesis advisor: Stephen Kaplan Advisor's department: Psychology Thesis advisor: Paul D. Scott Advisor's department: CCS Committee member: Michael D. Gordon Member's department: Mgmt. Info. Systems - Business School Committee member: John H. Holland Member's department: CCS Committee member: Robert K. Lindsay Member's department: Psychology Research institution: Univ. California - San Diego Computer Science & Engr. Dept. Inst. location La Jolla, CA Dates: 9/1/86 - present ------------------------------ End of Neurons Digest *********************