neuron-request@HPLABS.HP.COM ("Neuron-Digest Moderator Peter Marvit") (01/16/89)
Neuron Digest Sunday, 15 Jan 1989 Volume 5 : Issue 4 Today's Topics: Copies of DARPA Request for Proposals Available Human Learning & Connectionist Models Re: INNS application Re: INNS application medical applications of computer neural networks Neural nets for spatial reasoning? Neural Networks in Natural and Artificial Vision PDP Vol III simulator on a MAC? Post-processing of neural net output Re: Post-processing of neural net output Second SIMILARITY METRICS Posting Submission-Neural Learning Methods Re: talk at ICSI VLSI Implementations of Neural Networks Send submissions, questions, address maintenance and requests for old issues to "neuron-request@hplabs.hp.com" or "{any backbone,uunet}!hplabs!neuron-request" ARPANET users can get old issues via ftp from hplpm.hpl.hp.com (15.255.16.205). ------------------------------------------------------------ Subject: Copies of DARPA Request for Proposals Available From: will@ida.org (Craig Will) Date: Tue, 03 Jan 89 10:50:14 -0500 Copies of DARPA Request for Proposals Available Copies of the DARPA Neural Network Request for Proposals are now available (free) upon request. This is the same text as that published December 16 in the Commerce Business Daily, but reformatted and with bigger type for easier reading. This version was sent as a 4-page "Special supplementary issue" to subscribers of Neural Network Review in the United States. To get a copy mailed to you, send your US postal address to either: Michele Clouse clouse@ida.org (milnet) or: Neural Network Review P. O. Box 427 Dunn Loring, VA 22027 ------------------------------ Subject: Human Learning & Connectionist Models From: gluck@psych.Stanford.EDU (Mark Gluck) Date: Thu, 05 Jan 89 07:17:32 -0800 [[Editor's note: I hope Mark summarizes his responses and sends them in. I, for one, would be very interested in the results. -PM]] I would grateful to receive information about people using connectionist/ neural-net approaches within cognitive psychology to model human learning and memory data. Citations to published work, information about work in progress, and copies of reprints or preprints would be most welcome and appreciated. Mark Gluck Dept. of Psychology Jordan Hall; Bldg. 420 Stanford University Stanford, CA 94305 (415) 725-2434 gluck@psych.stanford.edu. ------------------------------ Subject: Re: INNS application From: lehr@isl.Stanford.EDU (Michael Lehr) Organization: Stanford University Date: 01 Jan 89 02:52:37 +0000 [[Editor's note: This relates to a series of messages last fall about INNS membership and the apparent lack of response from the organization. I hope the controversy will be resolved herein. -PM]] In article <19202@shemp.CS.UCLA.EDU> rosen@CS.UCLA.EDU () writes: >Although I signed up (and paid) for my INNS membership last September in >Boston, I have yet to receive my first (or any) issue of the Neural Network >journal, yet I have just received notice to renew my membership. >Does anyone know when (or, at this point if) they will mail the journal >to those of us who have paid for it? I feel somewhat foolish renewing >my membership and paying my dues at this point - I wonder just what I >am paying for. > Bruce >Bruce Rosen > ARPA: rosen@CS.UCLA.EDU > UUCP: ...!ucbvax!ucla-cs!rosen Dr. Widrow asked me to relay this message in response the journal problem: I have received several complaints from INNS members about not receiving their copies of the journal NEURAL NETWORKS. If this has happened to you, please accept my apology. The person who is now in charge of membership is: Frank Polkinghorn 9202 Ivanhoe Road Ft. Washington, MD 20744 Frank is working hard to allieviate the logjam. I am pleased to report that INNS has grown from zero to 4000 members in a little over one year. We are having some "growing pains." If you do not receive you journals within about a month, write to Frank and let him know the facts of your situation. I am sure that he will get your journals to you soon. Sincerely, Bernard Widrow President INNS -ml ------------------------------ Subject: Re: INNS application Organization: The Santa Cruz Operation, Inc. Date: 06 Jan 89 20:56:08 +0000 [[Editor's note: Well, at least one happy customer! -PM]] >>In article <19202@shemp.CS.UCLA.EDU> rosen@CS.UCLA.EDU () writes: >>I paid also in Sept. --- just received by renewal notice --- yet I have >>never received the INNS journal. I paid in September, heard nothing until November, and then received Vol. 1, No. 4 from Pergamon Press. Then, just yesterday, Vol. 1, Nos. 1-3 arrived, as well as the abstracts from the September INNS meeting held in Boston. So now I'll probably renew. FYI, here's the address for Pergamon Press: Pergamon Journals, Inc. Fairview Park Elmsford NY 10523 ------------------------------ Subject: medical applications of computer neural networks From: sid@koko.UUCP (Dave Sidney) Organization: Calif. State Univ., Stanislaus, Turlock, Ca Date: 29 Dec 88 09:44:14 +0000 Anybody have any information on current medical research or applications using computer neural networks? What is being done where, by whom, and with what results? ------------------------------ Subject: Neural nets for spatial reasoning? From: engelson-sean@CS.YALE.EDU (Sean Philip Engelson) Organization: Yale University, New Haven, CT Date: 04 Jan 89 03:51:04 +0000 I'm looking for references on neural-net models of spatial reasoning tasks. Any sort of spatial reasoning, learning or retrieval tasks involving connectionist architectures would be appreciated. Thanks, -Sean- Sean Philip Engelson, Gradual Student Mi ha'ish hachafetz chayim, Yale Department of Computer Science Ohev yamim, lir'ot tov? 51 Prospect St. Sur mera` v`aseh tov, New Haven, CT 06520 Bakesh shalom, v'rodfehu! ------------------------------ Subject: Neural Networks in Natural and Artificial Vision From: daugman%charybdis@harvard.harvard.edu (j daugman) Date: Fri, 06 Jan 89 10:41:42 -0500 For preparation of 1989 conference tutorials and reviews, I would be grateful to receive any available p\reprints reporting research on neural network models of human / biological vision and applications in artificial vision. Thanks in advance. John Daugman Harvard University 950 William James Hall Cambridge, Mass. 02138 ------------------------------ Subject: PDP Vol III simulator on a MAC? From: Keith Stenning <keith%epistemi.edinburgh.ac.uk@NSS.Cs.Ucl.AC.UK> Date: Sun, 15 Jan 89 14:34:56 +0000 [[ Editor's note: This sems to be a periodic request, but I still haven't heard of any source. Can a reader help out here? -PM ]] I am looking for a version of the CMU PDP Vol. 3 package of PDP simulations converted for the MAC. That's the package that comes with McClelland and Rumelhart Vol III. Can anyone help? Keith Stenning (keith@epistemi.ed.ac.uk) ------------------------------ Subject: Post-processing of neural net output From: mesard@BBN.COM Date: Wed, 28 Dec 88 16:40:39 -0500 A large portion of existing feedforward nets take an input vector and produce an encoded boolean response on their outputs (e.g., yes/no, left/right, signal/noise). Typically, the output is interpreted based on the activation of a singleton output unit exceeding some threshold, or whether the activation of one unit is larger than another. This approach may mean throwing away a lot of information. Even if the activation fails to meet some criterion, it might be useful as a similarity measure, or a rating of the net's confidence in its output. Furthermore, the criterion could be adjusted based on the discriminablity between outputs in the positive and negative cases. Does anyone know of any work that has been done in this area? void Wayne_Mesard(); Mesard@BBN.COM Bolt Beranek and Newman, Cambridge, MA ------------------------------ Subject: Re: Post-processing of neural net output From: terry@cs.jhu.edu (Terry Sejnowski <terry@cs.jhu.edu>) Date: Fri, 30 Dec 88 16:32:32 -0500 The value of an output unit is highly correlated with the confidence of a binary categorization. In our study of predicting protein secondary structure (Qian and Sejnowski, J. Molec. Biol., 202, 865-884) we have trained a network to perform a three-way classification. Recently we have found that the real value of the output unit is highly correlated with the probability of correct classification of new, testing sequences. Thus, 25% of the sequences could be predicted correctly with 80% or greater probability even though the average performance on the training set was only 64%. The highest value among the output units is also highly correlated with the difference between the largest and second largest values. We are preparing a paper for publication on these results. Terry Sejnowski ------------------------------ Subject: Second SIMILARITY METRICS Posting From: king@rd1632.Dayton.NCR.COM (James King) Organization: R&D, NCR Corp., Dayton, Ohio Date: 12 Jan 89 15:46:57 +0000 *** SECOND POSTING ON SIMILARITY METRICS *** About a month ago I posted a request for information, and interest, in the area of "similarity metrics". I am posting a second call for information now with the hope of furthering my base of understanding, and also to develop a base for the semi-formal survey I will be sending all respondents (this will be out in a week or so). I have received feedback from about thirty people. Most of the respondents have described their interests in this area, and many have provided abstractions of their methods for measuring similarity. This is very encouraging and I hope it continues. The survey will also be sent to a sizeable number of researchers that I know of already. My hope is to make this a cross-discipline study that provides insight from the Case-Based Reasoning, Analogical Reasoning, EBL, Information Retrieval, Behavioral Studies, Machine Learning, Child Psychology, etc. fields. At present a proposal for holding a workshop on this topic at IJCAI has been decided against. The topic may be presented as a panel discussion at a focused Case-Based Reasoning Workshop this year. --------------------- Original (edited) posting follows ------------------- SIMILARITY ... What does it mean? for ANALOGY What are the measures? for REMINDING Are there generalities or is it domain-specific? for EXEMPLARS ... Eliciation strategies, cues, weights, features, etc. I am performing independent research in the area of Case-Based Reasoning, CBR, and I am working on various metrics for similarity. In general, what ideas do you (the net-world) have about: - What about a new situation reminds you of a prior experience? - OR - How does one situation remind you of another? A little more focus might be how does one discriminate and weigh features of a new situation (case) in relationship to a large case-base of experiences that may or may not have a bearing on the new situation. Did that provide more focus or fuzziness!? This notice is sent out as a preliminary "attention-getter" to provide myself with some input to help form a more formal survey. Once written I hope to send it to a specific set of researchers (consisting mostly of people in the CBR, information retrieval (IR), doc. mngt. areas) and to anyone in netland that requests so. If anyone is interested in responding to any of this: - I will watch the "nets" for replies - Email to: j.a.king@dayton.ncr.com - Call: (513)-445-1090 before 4:30 (EST) (317)-478-5910 after 6:00 - Mail: NCR Corp. 1700 S. Patterson WHQ-5E Dayton, OH 45479 The survey will be finished and sent in the next week or two, so please let me know of your interest and what YOU might like to get out of such a study. Thank you for your time. Jim King ------------------------------ Subject: Submission-Neural Learning Methods From: David Kanecki <kanecki@vacs.uwp.wisc.edu> Date: Mon, 09 Jan 89 20:02:07 -0600 I am interested in studying the various learning rules used on neural networks. Can anyone send me a re article reference or tutorial on the following: First Order Delta Rule Second Order Delta Rule First Order Methods or Second Order Methods a. Delta Rule b. Hebb Rule c. Gradient Decent d. Back Propagation e. Kohonen's rule f. A.H. Klopf and B. Kosco rule g. Others not named above In exchange I will compile a list and send it to the Neuron Digest. Thank you, David Kanecki,ACS/Bio. Sci. ------------------------------ Subject: Re: talk at ICSI From: baker%icsi.Berkeley.EDU@berkeley.edu (Paula Ann Baker) Date: Thu, 05 Jan 89 08:44:19 -0800 [[ Editor's note: I got thisnotice and then lost it. I did make it to the talk, though, which was quite interesting. Dr. Baggi, a music professor, gave some very nice aural demonstrations of "computer-generated" swing, but was a little short on explaining the actual connectionist implementation. His approach would be best described as classic expert-system style, in which he hand fed the possibilities and structures and used the simulator for not much more than "Mozartian dice." So far, he seems to have found no particular advantage to the connectionist architecture, apart of the ease of programming "paralell" events. Training the network to analyze swing and *then* synthesize new music would be most intriguing, but he said that he wouldn't work on that any time soon. -PM ]] The International Computer Science Institute is pleased to present a talk: Wednesday, January 4, 1989 12:00 p.m. "A Connectionist Model for the Synthesis of Swing in Afro-American Jazz" Dr. Denis Baggi University of California, Berkeley The introduction of computer technology to problems of music and musicology is as old as computer science and goes back to Charles Babbage and Ada Lovelace. Starting with the Illiac Suite in 1956, considerable progress has been achieved both in computer assisted sound synthesis and in algorithms for symbol manipulation in composition and music analysis. Several centers - Bell Labs., CCRMA at Stanford, IRCAM in Paris, the Experimental Music Studio at MIT - have been dedicated to these problems. With less than half-a-dozen exceptions, however, no results of applications of computer techniques have been obtained for the psychoacoustical, perceptual problem of swing in Afro-American Jazz. Properly, and holistically, defined, swing is the medium of the Jazz message - as space is the medium of sculpting. Technically, swing consists of patterns of rhythmic accentuations and tension-release acoustical devices which are almost instinctively perceived by the listener - though an analytic description, let alone a teaching methodology, is elusive. This talk describes some work in progress for the construction of a connectionist model for the synthesis of swing. Using the Rochester Connectionist Simulator at the International Computer Science Institute in Berkeley, a system is being developed which accepts as input the harmonic grid of a Jazz standard and which constructs the lines played by a rhythm section consisting of piano, bass and drums. The main program is built around a connectionist network consisting of several automata running in parallel such that control passes from one to the other. The output drives a digital sound synthesis machine. Some preliminary results will be played and discussed. This talk will be held in the Main Lecture Hall at ICSI. 1947 Center Street, Suite 600, Berkeley, CA 94704 (On Center Street between Milvia and Martin Luther King Jr. Way) ------------------------------ Subject: VLSI Implementations of Neural Networks From: josh@flash.bellcore.com (Joshua Alspector) Date: Fri, 06 Jan 89 14:32:55 -0500 I will be giving a tuturial on the above topic at the Custom Integrated Circuits Conference. Vu grafs are due at the end of February and I would like to include as complete a description as possible of current efforts in the VLSI implementation of neural networks. I would appreciate receiving any preprints or hard copies of vu grafs regarding any work you are doing. E-mail reports are also acceptable. Please send to: Joshua Alspector Bellcore, MRE 2E-378 445 South St. Morristown, NJ 07960-1910 ------------------------------ End of Neurons Digest *********************