neuron-request@HPLMS2.HPL.HP.COM ("Neuron-Digest Moderator Peter Marvit") (12/12/90)
Neuron Digest Tuesday, 11 Dec 1990 Volume 6 : Issue 70 Today's Topics: Astronomy and Medical Imaging Applications? neuron digest item job opening - Rank open NEURAL NETS COMPUTING WITH GAUSS Re: Transputer Implementations References? standardization Help with new HITACHI product Academic Program Info Request post-doctoral position in neural nets in France "Request for posting a message" Neural net who plays chess Re: Help with new HITACHI product - SUMMARY of the information Japanese enter nets computer business - FYI Neural Chess - Questions, Comments and Options Re: The MIC and the Individual to creativity and reward Re: Articles on fuzzy cognitive maps & NN predictive modeling Send submissions, questions, address maintenance and requests for old issues to "neuron-request@hplabs.hp.com" or "{any backbone,uunet}!hplabs!neuron-request" Use "ftp" to get old issues from hplpm.hpl.hp.com (15.255.176.205). ------------------------------------------------------------ Subject: Astronomy and Medical Imaging Applications? From: Ade Miller <ASM%ASTRONOMY.PHYSICS.SOUTHAMPTON.AC.UK@pucc.PRINCETON.EDU> Date: Tue, 27 Nov 90 15:03:00 +0000 Having just completed a fairly extensive search for network applications in astronomy and medical imaging I was surprised to find only on or two papers in each field. Is anyone out there doing anything in either of these fields, on any sort of hardware? Thanks, Ade. ------------------------------ Subject: neuron digest item From: codelab@psych.purdue.edu (Coding Lab Wasserman) Date: Tue, 27 Nov 90 15:23:09 -0500 BRAINS' BRAINS Landman offered the comment that Gauss' brain was noticeably more convoluted than normal. However, striking observations of this sort are an inevitable concomitant of the variation in human brains. By the same token, observations in the other direction are not difficult to come by. For example, I believe, without having the citation ready to hand, that the brain of Anatole France was only about 2/3 average size. There is great variation in the size and shape of both genetically unselected and selected brains. Wild (i.e., unselected) brains vary as much from one person (or primate) to another as do faces. As Wahlsten has shown, even inbred strains exhibit substantial gross variations in morphology (e.g., about 20% of one inbred strain of mice do not develop a corpus callosum while the rest of that same strain do develop one, albeit to varying extents). So the noise in any gross brain size measurement is enormous. Is there any signal about individual differences that can be extracted from this noise? Well, the phrenologists tried to prove that there was, and they spent about a hundred years of well-financed research time in that attempt without any lasting success. Therefore, there is probably not much doing here. Jerry Wasserman Purdue University ------------------------------ Subject: job opening - Rank open From: BOHANNON%BUTLERU.BITNET@VMA.CC.CMU.EDU Date: Thu, 29 Nov 90 11:59:00 -0500 Applied Cognition - Rank Open: The Department of Psychology at Butler University is seeking nominations/applications for a tenure track opening to start August, 1991. We are seeking faculty who would be knowledgeable in distributed processing systems and their applications. Specific area is less important than evidence of excellence in both teaching and research. Salaries are negotiable and competitive. Responsibilities include: maintain an active research program with undergraduates, teach an undergraduate course in Human Factors and other courses in the candidate's area of interest. Minimum Qualifications: Ph.D. in psychology at time of appointment. For junior nominees/applicants: potential excellence in both research and teaching. Teaching experience preferred. For senior nominees/applicants: established record of excellence in research, more than 3 years of teaching experience, and an interest and/or experience in attracting external funding. Butler University is a selective, private university located on a 400+ acre campus in the residential heart of Indianapolis. The psychology department is housed in recently renovated space with state-of-the-art, video/social interaction and computer-cognition labs (Macintosh II and up). Screening of nominees/applicants will continue until suitable faculty are found. Those interested should send a statement of research and teaching interests, vita, and three original letters of reference to: John Neil Bohannon III, Head, Department of Psychology, Butler University, 4600 Sunset Ave., Indianapolis, IN 46208. Bitnet : Bohannon@ ButlerU. AA/EEO ------------------------------ Subject: NEURAL NETS COMPUTING WITH GAUSS From: zhou@brazil.psych.purdue.edu (Albert Zhou) Organization: Purdue University Date: 02 Dec 90 01:31:02 +0000 We've been using a programming language GAUSS to do neural network computing, and we would like to share experiences and ideas with other neural nets workers using the same language. For those who haven't heard of it before, Gauss is a powerful high-level matrix-oriented language implemented in PC environment. ------------------------------ Subject: Re: Transputer Implementations References? From: j_millan@cen.jrc.it (Jose del R. MILLAN) Date: Mon, 03 Dec 90 16:57:37 +0000 In reply to Ade Miller, Subject: Transputer Implementations References? > ... My main interests are back- propagation networks; optimisation > and learning, applications to the above and implementations on > parallel hardware (especially Transputers). ... Millan, J. del R. & Bofill, P. (1989). Learning by back-propagation: a systolic algorithm and its transputer implementation. The Int. Journal of Neural Networks: Research & Applications, 1(3), 119--137. Bofill, P., Manyer, J., Millan, J. del R., & Salvans, V. (1990). A systolic algorithm for back-propagation: mapping onto a transputer network. In Stender, J. & Addis, T. (eds.) "Symbols versus Neurons?". IOS Press. Jose del R. MILLAN Institute for System Engineering and Informatics Commission of the European Communities. Joint Research Centre Building A36. 21020 ISPRA (VA). ITALY j_millan@cen.jrc.it (bitnet) ------------------------------ Subject: standardization From: atilla gunhan <gunhan@cc.uib.no> Date: Tue, 04 Dec 90 11:46:25 +0100 TERMINOLOGICAL AND METHODICAL STANDARDIZATION. The most significant problem which may occur during an educational program is the existence of different terminology and/or methodology used in the various literature of a given subject. Such problems has existed almost in any kind of subject at the Universities. In the Neural Network field, one can easily recognize such problems. Graphical representations, variables, concept definitions, classifications in the subject vary from book to an other. Therefore I am willing to write a paper in which I propose a standardized terminological and methodical approach for ANN. I would like to hear your opinions. E-mail: gunhan@cc.uib.no Address: Atilla Gunhan Dept. of Information Science University of Bergen Hightechnology center in Bergen Thormohlensgt. 55, N-5008 Bergen, Norway ------------------------------ Subject: Help with new HITACHI product From: Paulo V Rocha <P.Rocha@cs.ucl.ac.uk> Date: Wed, 05 Dec 90 08:59:16 +0000 The New York Times recently (sorry, I don't have the issue) reported on Hitachi's new Neurocomputer, which claims an execelent performance. Can anyone help me with some pointers to where I can find a description of the architecture (or details in general)? Maybe someone in Japan? I haven't heard anything about it here in the UK. Thanks and of course I will post the information I get :-) P. +-----------------------------+---------------------------------------------+ Paulo Valverde de L. P. Rocha | JANET:procha@uk.ac.ucl.cs Department of Computer Science| BITNET:procha%uk.ac.ucl.cs@UKACRL University College London |Internet:procha%cs.ucl.ac.uk@nsfnet-relay.ac.uk Gower Street | ARPANet:procha@cs.ucl.ac.uk London WC1E 6BT | UUCP:...!mcvax!ukc!ucl-cs!procha England | tel: +44 (071) 387 7050 x 3719 | fax: +44 (071) 387 1397 +-----------------------------+---------------------------------------------+ ------------------------------ Subject: Academic Program Info Request From: worth@park.bu.edu (Andrew J. Worth) Date: Wed, 05 Dec 90 11:28:44 -0500 ISSNNet Request for Academic Program Information The International Student Society for Neural Networks (ISSNNet) is compiling a list of academic programs relating to Neural Networks from around the world. We would like your input if you are a member of a scholastic program that is in any way related to Neural Networks, Parallel Distributed Processing, Connectionism, Computational Neuroscience, Neural Modeling, Neural Computing, etc. We hope to provide this service so that (1) interested students will be able to apply to those programs that will most closely satisfy their educational goals, and (2) current students and non-students will be aware of existing academic programs. This service is intended to not only provide an overview of these programs and contact points for more information, but also a personal glimpse into what's behind the official descriptions. All information will be made publicly available and will be updated as new programs are created and as programs change. Complying with ISSNNet's goal to be absolutely unbiased, we would like this to become THE source of information on academic programs in this field. ISSNNet would like to provide the following information: - Official address to contact for more information (surface mail and email) - Official description of the program. - Names of Faculty Members and their interests - Degrees requirements (BA, BS, MA, MS, PhD, etc.) - Short description of courses offered - Computing resources (Hardware and Software Tools) - Number of Students (grad/undergrad) and related faculty - A brief *personal* description of the program, department, etc. describing motivation, emphasis, goals, and/or overall ambiance. - Student Contacts (w/ telephone numbers, email and surface addresses, degree sought, interests, and date of graduation) This information is above and beyond the academic questionnaires that were filled out at the San Diego and Paris conferences and will eventually be made available via ftp and also by other means through ISSNNet (your submission will be taken as permission to make the information public unless we are otherwise notified). Coordinated responses from each institution are encouraged and will be appreciated. Please submit descriptions of academic programs in plain text (email is preferred) following the guidelines above to: issnnet-acad-progs@bucasb.bu.edu We will also be able to re-distribute information in other emailable formats such as postscript or LaTeX. Thank you for your time and effort, Andy. ---------------------------------------------------------------------- Andrew J. Worth (617) 353-6741 ISSNNet ISSNNet Academic Program Editor P.O. Box 557 issnnet-acad-progs@park.bu.edu New Town Br. worth@park.bu.edu Boston, MA 02215 USA ------------------------------ Subject: post-doctoral position in neural nets in France From: "James A. Reggia" <reggia@cs.UMD.EDU> Date: Thu, 06 Dec 90 14:33:00 -0500 Post-Doctorate Position in France: A two year post-doctoral position in neural modelling is available at ONERA/CERT in Toulouse, France. ONERA/CERT is a government research laboratory: Office National d'Etudes et de Recherches Aerospatiales/Centre d'Etudes et de Recherches de Toulouse. The pay is approximately $2000/month. The working language is French, but most individuals at ONERA/CERT speak English fairly well. Work in this position would include one or more of the following: development and study of learning rules in competitive systems, matching images, or development of neural modelling software on a connection machine (the latter would require spending some time in Paris too). For further information or for answers to questions, please contact Paul Bourret at bourret@tls-cs.cert.fr via email. ------------------------------ Subject: "Request for posting a message" From: VEMURI@icdc.llnl.gov Date: Thu, 06 Dec 90 14:45:00 -0700 Dear friends: I am interested to know if any one of you, out there, is working on the application of artificial neural nets to the drug abuse problem. The range of potential problems is broad: 3_D drug design, data bases for molecules, brain/reward mechanisms, and so on. Thanks. Rao Vemuri vemuri@icdc.llnl.gov Dept. of Applied Science. ------------------------------ Subject: Neural net who plays chess From: morphy@cobalt.cco.caltech.edu (Jones Maxime Murphy) Date: Thu, 06 Dec 90 16:33:01 -0800 I am a chess master, and I have actually done some work on building a net to solve K+P endgames. For a master, K+P endings are quite easy to evaluate, and I also had the K+P evaluator from "Cray Blitz", which I modified slightly for use as a training tool for my networks. I concentrated on the K+PvK endgame, the simplest possible endgame. My 6 inputs consisted of the position of the pawn and the vectors from the pawn to the kings. The output was to be 1 if the position was a win, or 0 if the position was a draw. I generated every possible position and trained the network on the output from "Cray Blitz"'s evaluator. "Cray Blitz" is a well-known computer program which competes running on Cray machines. The first problem was that the Cray Blitz evaluator sucked. The person who wrote that thing must not have been much of a player. I spent lots of time hacking their code. Then, when I finally got it to the point where I was halfway satisfied with it, I ran into the real problems. First of all, edge effects. The edges of the chessboard have a really major effect, since translation invariance only holds for 6 files. This means a substantial proportion of postions are rookpawns, which are special. A really helpful suggestion from Christof Koch, which I hope to implement, is to make the board horizontally infinite. This will mean monkeying around with Cray Blitz(gag!). As to the nice idea of alpha-beta/neural net duality, I(and others, I'm sure) have been pondering this since 1986. The computational price is extremely heavy, as I found out with my little K&P thing. One training run would take 18 hrs on a Sun 3. I hope that work continues, tho. And, sorry, I can't shed much light on how masters process positions. Friends of mine who are much stronger players than I are often at a loss to articulate how or why they make certain key moves. It's something that we do talk about a great deal, though. Jones ------------------------------ Subject: Re: Help with new HITACHI product - SUMMARY of the information From: Paulo V Rocha <P.Rocha@cs.ucl.ac.uk> Date: Fri, 07 Dec 90 09:06:05 +0000 The most recent publication about the circuit seems to be "Design, Fabrication and Evaluation of a 5-inch Wafer Scale Neural Network LSI Composed of 576 Digital Neurons" by M.Yasunaga, et.al., Proc.IJCNN'90(San Diego) pp.II-527. I also received some details from a pamphlet probably distributed at Hitachi's technology fair in New York. - - ------------------ Neuron circuit: Completely digital circuit with learning function. Architecture: Time sharing digital bus Dual networks for learning Performance of learing circuits: 2.3 GCUPS Neumber of neurons: 144 neurons per wafer 1152 neurons per system Neuron output: 9 bits Synapse weight: 16 bits Process: 0.8 micron CMOS gate array Wafer size: 5" diameter System size: 30 cm x 21 cm x 23 cm Power consumption: ~50 Watt - - --------------- The circuit uses a single bus and the claimed performance is a peak figure of operational units (multipliers and adders). Real performance can be much less. Each neuron has only 64 synapses because the the design uses gate array technology and memory in this case is very expensive. The neurocomputer has about 70 K synapses and 1 K neurons. Thanks again to everybody. Paulo (University College London) ------------------------------ Subject: Japanese enter nets computer business - FYI From: havener@Kodak.COM Date: Sat, 08 Dec 90 11:01:03 -0500 John, FYI. DEC may miss the boat on this business if they don't get something underway. ============================================================================== Note 313.0 Hitachi neurocomputer announced No replies TXTC01::TXTC01 60 lines 7-DEC-1990 02:01 According to an article in the November 26th edition of the "Nikkei Shinbun" (the Japanese equivalent to the Wall Street Journal), Hitachi has developed the world's fastest general purpose neurocomputer. The learning speed is reported as 2,300 Million Weight Updates Per Second (MWUPS), approximately four times that of the Fujitsu neurocomputer which had a reported 587 MWUPS. It is constructed of eight 5 inch silicon wafers enclosed within a 30cm x 21cm x 23cm cabinet. Each wafer contains 144 neurons, making the total capacity 1,152 neurons. It is designed to be used in conjunction with a workstation. In the same article, two applications were announced. The first is for stock value prediction and the second is for signature verification. The stock value prediction application appears to accept the last 20 days price data for a given stock as input, and output a projection for the next 10 days. The signature verification is trained on 5 actual signatures, and verification is made based upon the pen pressure, and horizontal and vertical pen speeds. The company plans to commercialize the neurocomputer withing the next two years at a price of less than 100 million yen (about $770k at $1 = 130). Related articles: %A Kato, H. %A Yoshizawa, H %A Iciki, H. %A Asakawa, K. %T A Parallel Neurocomputer Architecture with Ring Registers %J Proceedings of the InfoJapan '90 Computer Conference %V 1 %P 233-240 %I Information Processing Society of Japan %C Tokyo %D 1990 %K Neurocomputer %A Herbst, K. %T Fujitsu Parallel Design Speeds Learning of Neurocomuters %J Supercomputing Review %V 3 %P 56-57 %I Supercomputing Review %C San Diego %D October 1990 %K Neurocomputer - ---Aggies-O-Aggies!--We fight until we die!--We fight until we die!---- John P. Havener / / Senior Chemical Engineer / / / / TEX Core Artificial Intelligence Group ------------- Eastman Kodak, Texas Eastman Co. ---/ New Mexico /--- Building 1, Box 7444 -------/ State /------- Longview, Texas 75607 -------/ Aggies /------- Phone: (903)237-6368 ---/ 1978 /--- Net: havener@Kodak.com -------------- Fax: (903)237-5371 / / / / / / ---Aggies-O-Aggies!--We'll win this game and that's the reason why!------- ------------------------------ Subject: Neural Chess - Questions, Comments and Options From: David Kanecki <kanecki@vacs.uwp.wisc.edu> Date: Sat, 08 Dec 90 20:19:35 -0600 Recently I was asked the following questions about the neural chess program: 1) What is the architecture of mamalian neural net? 2) What method is used - Backprop, cascade, etc? 3) What input units are used - whole board, single unit? 4) What type of training protocols are used? 5) How does the computer choose a move? 6) Define 'minimalist learning'? 7) Hows does the concept of 'Decision Making' apply to neural chess? This information is considered propietory system programming which will be available on published papers in the future as I am enrolled in pre-doctoral program in biological science. Some people have asked if there was a way they could see or play against the program. Yes, they can. To see the program I can prepare a VHS tape of selected moves of the computer screen of the match from start to completion. Upon review of this tape I would like to receive comments or suggestions. Send $10 for shipping and handling and upon return of the tape I will refund the tape fee. To play the program I suggest the following: 1) The person would need to send me either a board position that they would wish to start from. 2) Or, they could play chess againstthe program. Both options are available via correspondence where as a move(s) is received, the output will be returned through e-mail. 3) VHS tape option as described above. The primary reason for announcing the program was to show that there are independent creative persons who can share what they have developed but others are who are working may not be able to report their finding. However, your comments and feedback to my announcement will be of value to all interested in this area. This program was developed without grants or funding. The development process includes 10 years of computer and biological science education in university disciplines. Any comments or feedback will be updated through e-mail as well as any published papers when they are available. Finally, I would encourage all to suggest comments or suggestions of pending work or work they wish could be done so that others would be enriched and encouraged "remember, that this program was developed on a 64K CPM computer through wise programming options and selections". More people should be encouraged to write programs on a 64K basis before going to higher capacity machines in order to develop their minds. Similarly, I have developed a neural simulation program on a 64K basis that was reported earlier. David H. Kanecki, Bio. Sci., A.C.S. P.O. Box 93 Kenosha, WI 53141 (414)-654-8710 kanecki@vacs.uwp.wisc.edu ------------------------------ Subject: Re: The MIC and the Individual to creativity and reward From: David Kanecki <kanecki@vacs.uwp.wisc.edu> Date: Sat, 08 Dec 90 21:15:21 -0600 [[ Editor's Note: Once again, I fear this message strays into an area best left to other fora. However, I'm letting it go and will entertain one round of replies, if any readers wish to do so. Remember, this should be primarily a *technical* Digest relating to neural networks, both natural and artificial. Again, I tend to be quite liberal in my editorship but feel this is getting a bit afield. -PM ]] One of the last speeches by the late President Eisenhower was on the buildup of the Military Industrial Complex and the result upon the individual. His fear was that the MIC would acquire most of the funding with the individual being forced to give their idea to a MIC organization whereby the MIC would receive recognition and credit and individual nothing. And, the individual would have to play ball. The point of my neural chess program is that there are creative people who prefer not to work for that system. Also, I think it would be better if funding and credit were awarded to the individual. Also, the merit of the idea should be the basis for funding not the education level or the number of papers one has written or co-written. Doctorial degrees were present only after the mid 1800's. Many people who developed and invented new ideas, concepts, and products had various types of education and skills. Also, all of them had the drive and ability to know that there is something better. One example of an organization receiving credit and funding and the individual nothing is the case of my father. He worked and was creative member in the research and development of the carbon rocket nozzle used in NASA Projects Mercury, Gemini, and Apollo. He also worked in other companies setting up advanced engineering departments/ projects. But, when he reached the age of 45 the company rewarded him by letting him go without a pension or reward. In fact he had to survive for 15 years using his small savings to support his family and send his children to college. Also, when he attempted to find other employment he had suddenly became too old and his past achievements were not valued. Is this fair or just to the remaining elderly professionals who are still living and could contribute but are not allowed to? What can we do to regain this technology before it is too late and lost? Is individual reward and creativity the thing that should be encourage while time is of essence! "Through mentors, grants, etc" I believe that if the U.S. wants to achieve the level of creativity it had during the space program it should provide support for the individual and not the organization. By providing too much support to an organization one would have the problems that now exist with the shuttle program, NASA, and other areas. Funding and credit should be given to individuals in all fields on the merit of their idea as opposed to their title or pull. These opions are my own. But, I thought is was time that someone addressed these issues. Any pros or cons comment would help strengthen the professional development community and the special interest areas covered by e-mail. In the key facts of success the first point is a change in attitude which is addressed above and the second point is pareto analysis where only 5 or 10% are the key issues that relate to over 80% of the needs to be done. David Kanecki, Bio. Sci., A.C.S. P.O. Box 93 Kenosha, WI 53141 kanecki@vacs.uwp.wisc.edu ------------------------------ Subject: Re: Articles on fuzzy cognitive maps & NN predictive modeling From: Willie Brown <brown@Corona.ITD.MsState.Edu> Date: Mon, 10 Dec 90 14:36:47 -0600 I'm looking for articles on FCMs & using neural networks as predictive models. I have copies of Bart Kosko's "Adaptive Inference in Fuzzy Knowledge Networks" and "Fuzzy Cognitive Maps" articles. Books and periodicals are also of interest. I'm looking at applying FCMs to predict the future availability of obsolete military microcircuits. Willie C. Brown Computer Engineer Center for Military Replacement Parts Starkville,MS ------------------------------ End of Neuron Digest [Volume 6 Issue 70] ****************************************