|Volume 1: No. 01|
Send me a message if you need to follow up on anything you read here. I can at least identify a contact person.
Barbara Simons, ACM National Secretary, reports that ACM committees are looking for input or participants in: design of ACM services and computer-science curricula to increase tech transfer; contacts and help for Eastern and Central European computer scientists; improvements in K-12 education via computer technology; and facilitation of non-sexist computer games.
Support for students giving presentations at neural-network conferences may be available from the International Student Society for Neural Networks (ISSNNet), or from matching grants by other societies. Contact email@example.com.
Tom Schwartz (The Schwartz Associates) quotes an MIT study of 700 neural-network researchers in 30 countries. 42% of the research had corporate funding, vs. 14.5% defense and 24.5% other government funding. Researchers came from diverse backgrounds: about 34% EE, 19% Physical Science, 18% CS, 7% Biology, 5% Cognitive Science, and 5% specifically Neural Network.
The new Computational Neural Network Centre (CONNECT) in Denmark is looking for people. Four immediate openings (at varying institutions) are in biological sequence analysis, analog VLSI for neural networks, neural signal processing, and optical neural networks. Benny Lautrup (firstname.lastname@example.org).
The Connectionists list has been discussing the precision required for various neural-network algorithms. Most take 12 to 16 bits for learning, but can then be run on 6 to 8-bit hardware.
Robert Siminoff is distributing a FORTRAN 77 simulation of the human retina, in conjunction with publication of a journal paper.
Spectrum (8/90) reported that the Society for Computer Simulation has listed 200 programs in its $15 Directory of Simulation Software. To get the 1991 questionnaire, write to SCS, Dept. SP, Box 17900, San Diego, CA 92117-7900; (619) 277-3888.
Canon is opening an intelligent-peripherals R&D lab in Palo Alto, with salaries rumored to include $70K for new Ph.D.s (and three times that for the most experienced managers or scientists). NEC has a research lab in Princeton, NJ; Hitachi is opening R&D labs in Farmington Hills and San Francisco; and other Japanese companies have been opening U.S. labs and plants. Newsweek reports that ASCII Corp., Tokyo, is planning to open a computer/television lab in the U.S. Much of the recent activity seems to be in multimedia and man-machine interfaces. (Speaking of which, IEEE now publishes Multimedia Review -- quarterly since Spring 1990 -- and the almost-monthly Virtual Reality Report.)
NEC's nationwide Japanese network now carries a database from the Kinokuniya bookstore chain listing 260,000 people, 140,000 books, and 750,000 articles. Access costs $1.50 per minute. [From MicroTimes, quoting the Japan Marketing Group.] In the U.S., Lotus has backed off on plans to publish a CD-ROM database of marketing data even though the ROM did not identify personal addresses. (Identical marketing data remains available via mainframe batch processing.)
Microsoft will set up an Asian R&D center in Tokyo in another year, growing from 200 to 500 engineers. [Also from JMG, 400 Groveland Ave., Suite 312, Minneapolis, MN 55403; (612) 871-7889.]
Tom Schwartz mentioned, at an AI Forum meeting, that the Japanese legislature is considering allowing foreigners up to 50% of intellectual property rights on joint projects in Japan, reversing a long-standing policy of the Ministry of Finance.
Tom also points out that copyright protection for neural networks is useless since alternative weight sets are easily constructed. (Trade secret protection is equally useless since black-box networks can be used to train others.) Software patents are thus likely to be employed. (GTE has been granted a patent on "momentum" in neural-network backpropagation. Publications prior to patent application can invalidate such patent claims.)
The League for Programming Freedom is opposed to software patents and interface copyrights, which apparently endanger the spontaneity and code sharing of the creative-hacker lifestyle (as well as the innovative code that results). Examples: U.S. Patent 4,197,590 on use of exclusive-OR for writing a cursor to the screen, and Patent 4,398,249 for natural-order recalculation of spreadsheets. Contact email@example.com for information. (I consider their case overstated.) Or contact Richard Stallman, firstname.lastname@example.org, for a lively discussion. Other prominent members are Richard Gabriel, Gerald Sussman, and Mitch Kapor. Membership is $42 for the employed, $10.50 for students, $21 otherwise.
To copyright your software -- which the League does not oppose -- call (202) 707-9100 for the U.S. Copyright Office Hotline (101 Independence Ave. SE, Washington, DC 20559). The fee for filing Form TX is just $20.
AI Magazine (Winter '90) carried an ad for a new monthly, the Intelligent Systems Report, from the publishers of AIWeek and Neural Network News. A 3-month trial was offered for $25. [I am not affiliated, of course, and the Communique does not carry paid advertising.] The contact number is (404) 434-2187.
For the business side of neural networks, Ed Rosenfeld is marketing Intelligence, a $295 monthly. Ed also has the Neural Network Almanac for $395. Or you can order SEAI's compendium, Neural Network Applications and Products, for $325. Call The Schwartz Associates, (415) 965-4561, for literature.
IEEE Spectrum (2/91) printed an ad for The Electronics Classifieds, a biweekly job-posting newspaper for electrical engineers; (602) 860-9535, price unspecified. Related services are Jobs-On-Line, (415) 324-3780 modem 8-N-1, and Execu-Net Information Services, (415) 947-6845. InfoDiscs (Palo Alto) markets a $40 DOS database of 1600 profiles for science and engineering companies in Silicon Valley; (415) 493-2212.
IEEE also says that Lederberg's survey/report on the state of grant funding in the U.S. is available from The American Association for the Advancement of Science, Directorate for Science and Policy Programs, 1333 H Street, Washington, DC 20005. (If you're planning to make a living on government grants, you may want to read the comments of frustrated respondents. NSF has published similar reports. Only 25% or so of NSF applicants win grants, with doubtful continuity from year to year.)
Helen Gigley (Program Director, KMCS) tells me that NSF has not continued its software initiative. Also, increased workloads in other proposal categories (due partly to the new "openness" policy) may delay both declinations and awards this year.
Cognitive scientists should take note of The Journal of Learning Sciences, a new quarterly from Lawrence Erlbaum Associates. Volume 1 (1991) is only $35 individual or $70 institutional -- $20 more outside the U.S. and Canada.
For information about government-supported robotics and AI projects -- including upcoming workshops and conferences -- government employees and contractors may access the free RAID database maintained by NOSC. Ask Mike Dwyer, (619) 553-5308, or CSC's Judi Graham, (619) 225-2511, about Milnet/DDN access.
If you're looking for a new AI customer, you might want to attend the First International Conference on AI Applications on Wall Street this October 9-11 in New York. Sponsors include ACM, IEEE, AAAI, and SMART-F$ (Society for the Management of AI Resources and Technology--Financial Services). Contact Mary Bianchi, (718) 260-3760, for information or registration.
To see what engineers have been up to, ask Pergamon Press for a sample issue of Engineering Applications of Artificial Intelligence. (The six issues for 1991 cost $355.) US: (914) 592-3359. UK and elsewhere: (+44) 0865-743479.
For information about the Macintosh world, you might consider MacInfo from Niles and Associates (Berkeley). The current disk should have about 7,000 abstracts from MacUser, Macweek, and other publications. $120 for 12 monthly updates ($99 for developers and academics). [If you'd rather network to get information, join a major users' group such as BMUG or BCS. I'm a BMUG member.]
GBA Inc. is compiling TechSpecs, a CD-ROM database of Macintosh product descriptions. Each listing is $295/year, plus $25 per CD ROM after the first quarter. (404) 518-1014.
Metatec Corp. is publishing the first monthly CD-ROM magazine, Nautilus, including classical and country music, games, photos, artwork, demonstrations, and educational material. (I couldn't determine the theme from Craig Crossman's article in the 12/2/90 San Jose Mercury News.) Subscription is $137.40; (800) 365-1639.
If this were a roundtable, we'd start by introducing ourselves. I'll be happy to introduce a few members each week if you will send me single-paragraph blurbs. I'm not going to introduce people without permission, since anyone who is job-hunting might prefer to remain anonymous. (Or might welcome the publicity.) If you do submit material or queries, I am likely to mention your name unless anonymity is requested.
In any case, I would appreciate receiving area-of-expertise or area-of-interest keywords. These will help me route news items to you that might not be relevant to the full membership. (I can also use my database of keywords to find members with needed domain expertise when I'm asked questions by others.) If you prefer not to have your mailbox filled with net clippings, of course, just let me know. I don't want to duplicate discussion lists that you already monitor.
In this issue, I would like to thank Harry Llull, Director of the Centennial Science & Engineering Library at UNew Mexico. He's helping me reach out to the library science/information science community -- an act of faith, given that he's never seen the Communique before today. (Harry is starting his own electronic journal/bboard for scientific and technical librarians, which should be interesting.) If others wish to help with membership, I'll be happy to send a stack of handouts that can be left at conferences. (It's a good way to recruit others with interests and expertise in your own field.) Or give me netnames of people that you'd like me to contact -- I'll mention your name.
I could also use some help in compiling address changes for influential people. If you happen note such an occurrence, please send it in. For example, I was told in November that Michael L. Baird of FMC has moved to Savoir (Hayward, CA) as VP and Director of Engineering and Technology. H. Glenn Haney of Dataquest (San Jose) was named CEO and Technology Division EVP. Guy Kawasaki -- Apple evangelist, ACIUS founder, and author/columnist -- is now with Salient Software. (Salient is developing Touch Base, a free-form contact/address database for personal or business networking.) Sanjay Mittal, who works in electronic document analysis, has moved from Xerox to Metaphor.
John Josephson (Ohio State) has asked me about the U.S. income-tax status of newsletters such as this. I'm not a tax expert, and I don't know whether tax issues differ for professional society memberships and for newsletters. However, there seem to be two bases on which to claim a deduction. An educational deduction may be claimed for training necessary to keep current in your own field (but not for training needed to switch fields). A business deduction may be claimed for any expense incurred to make income, or in reasonable expectation of profit. General-circulation magazines such as Inc. are usually not deductible, but financial newsletters may be if you use their advice to guide investments. Conference fees and professional-society dues are often deductible. Consultation with tax or legal professionals about business matters is almost always deductible. (The amount of help actually received is relatively unimportant.) Whether membership in Computists International is deductible would thus appear to depend on how you plan to use the service.
Computists International is a networking group. Such groups have long existed: churches, clubs, professional societies (or philosophical/Royal societies), fraternal orders, businessmen's associations. Computists is unique in that it uses global networks to serve information professionals. (I'd like to thank Stephen Wolff and NSF for permitting trial use of the NSFNet backbone in delivering this service to the research community.)
Career networking has itself become a topic of study over the past few decades. Many books claim that businessmen have always networked, and that businesswomen need to learn this skill. Clubs and associations for businesswomen now exist in every major city. My observation, however, is that scientists and engineers are just as much in need of such support -- especially the men, who tend to discuss only technical issues even during family get-togethers.
I now put forth the thesis that technical information is almost irrelevant to career success. It's not what you know, or even who you know, that's most important; it's how you think. The main reason for studying technical literature is to track who is doing what, where, paid for by whom. The other common reason for reading technical journals is to absorb the flavor: what sort of material is presented, how the presentation is structured, how the arguments are supported. Only very rarely does an experienced professional read a paper in order to implement the algorithms or replicate the experiments.
(I often start discussions by making outrageous claims. If you disagree, please jump in and say so. You won't offend me.)
My thesis holds just as strongly in the business world. A chief determinant of entrepreneurial success is having parents who were in business, or having business experience of your own at an early age. Why? Because you learn to think like a businessman. There are a hundred things you have to know, of course, but running one business is much like running another. Once you absorb the thought patterns, you can wing it. An MBA degree teaches little about day-to-day business, but gives you vocabulary and thought patterns so that you can quickly learn a particular business. (You can then become an advisor on business in general -- but beware the inexperienced expert.)
A technical focus, on the other hand, leads one into dependence on the system. You seem to be "climbing the ladder," but really are making yourself more and more interchangeable with others in the same technical specialty. This robs you of job security while doing little to open up managerial options. It does permit you to find work easily at other companies, but only if the economy is good, your specialty is currently hot, and you are willing to move. Even then, you are likely to be filling a predefined "slot" instead of doing work that really interests you.
You can't ignore the technical side, but you owe it to yourself to develop professional contacts and personal skills as well. At the minimum, talk to others about funding sources instead of algorithms. Standard networking advice is to make your net as diverse as possible -- and to keep it refreshed with frequent news clippings, queries, introductions, lunches, and Christmas/birthday cards. (Networks are like dynamic RAM; they fade out if not refreshed.) You never know when someone you know will know someone you need to know.
If your friends' net addresses have stopped working, drop me a note. I'm pretty good at figuring out modern equivalents for old internet/bitnet/uucp addresses. I can sometimes come up with street addresses or phone numbers as well. And if I can't figure one out, perhaps the other Computists will know.
I promised to keep the Communique short, so that's enough for this issue. Send me your news, interests, problems, and comments, and tune in next week.