|Volume 1: No. 07|
Rand Waltzman tells me that he's leaving DARPA the first of June for a tenured computer-science position at the Royal Institute of Technology (KTH -- Kunglige Tekniska Hogskolan) in Stockholm. His replacement as director of DARPA's image-understanding effort will be Oscar Firschein from SRI International (and, before that, from Lockheed). Steve Cross will continue to run DARPA's AI programs. Both jobs were formerly handled by Bob Simpson, who is now with NCR's Human Interface Technology Center in Atlanta.
If you'd like to congratulate Rand or Oscar, their net addresses are email@example.com and firstname.lastname@example.org. Oscar is a friend from my SRI days, when we studied Aikido together. (Don't mess with him!) He and Marty Fischler wrote a book on AI and vision a couple of years ago -- Eye and Brain -- stressing that perception is the most difficult and sophisticated level of intelligence. Oscar has long been involved with both aerial reconnaissance and the DARPA autonomous vehicles project. He also took over as head of SRI's planning and reasoning group when Michael Georgeff left to start the Australian AI Institute four years ago.
Craig Fields is trying to change MCC from an ivory tower into a cost-effective research house. Scientists, who had been "somewhat arrogant" about research schedules, must now stay in close touch with sponsoring organizations. Long-range research has been de-emphasized in favor of products nearly ready for marketing. MCC's mission is now "business success for our members," rather than long-term competitiveness.
Doug Lenat is reportedly unhappy at finding no sponsors for 5- to 10-year research in machine translation. Companies are looking for projects with a 90% guarantee of success. MCC claims transfer of 150 technologies to industry, but it has been too lean a return on the $372M spent to date.
Another development is the spin-out of entrepreneurial companies to exploit MCC's developments. Katherine Hammer's company, Evolutionary Technologies Inc., is the first. MCC owns a small piece, and corporate sponsors of the MCC research will get preferential rates on software licenses. [Evelyn Richards, Washington Post; from the San Jose Mercury, 5/5/91.]
Claudia Mazzetti, Executive Director of AAAI, will henceforth spend most of her time on long-term issues affecting the society. Carol Hamilton, Managing Director, will take over management of the AAAI office. [AI Magazine, Spring 1991.]
SIGAPP is a new ACM Special Interest Group on Applied Computing, including graphics, databases, communications, software engineering, AI, and office automation. The chair is Hal Berghel (email@example.com) of the Center for AI and Expert Systems, UArkansas. Other officers are George Hedrick (Oklahoma State), Beth Unger (Kansas State), and John Talburt (UArkansas at Little Rock). SIGAPP will assume responsibility for the 6th Symposium on Applied Computing and will publish a quarterly, Applied Computing Review. Contact ACM Headquarters to join. [ACMemberNet, 5/91.]
SIGLINK is a new ACM SIG for hypertext scientists, designers, and users. Officers include Robert Akscyn (firstname.lastname@example.org) of Knowledge Systems (Murrysville, PA), Janet Walker (DEC), Tim Oren (Apple), and John Legget (Texas A&M). [ACMemberNet, 5/91.]
Complete directories of academic e-mail conferences are now available via anonymous FTP from ksuvxa.kent.edu, or by sending a "GET ACADLIST FILEn" message to email@example.com. FILE1 and FILE2 cover Humanities and Social Sciences; FILE3 covers Biological/Medical Sciences; FILE4 is Physical Sciences; and FILE5 is a mix of Business and Academia (grants, academic freedom, etc.). These files will also be available in print format from the ARL in June 1991. A self-decompressing, binhexed Macintosh Word 4.0 version of the Directory in its entirety will be available on the FTP server soon. Contact Diane Kovacs (firstname.lastname@example.org or email@example.com, (216) 672-3045), Kent State Reference Librarian for the Humanities, with questions or updates. [LIBRES 1.4.]
In the last issue, I mentioned a CBR award to Cognitive Systems. Dan Corkill (firstname.lastname@example.org) notes that his Blackboard Technology Group, Inc., was awarded a Phase I SBIR entitled "Mixed-Paradigm Case-Based Reasoning using a Blackboard Architecture." A Phase I final report will be issued to DARPA in late September 1991.
Dan's project concerns control and problem solving in knowledge-based applications where CBR alone is insufficient. His GBB blackboard architecture integrates knowledge sources operating with CBR and other reasoning techniques. It seems that CBR and the blackboard paradigm share commonalities in indexing and retrieval (of cases and blackboard objects) and in large-grained control decisions (case selection and KS execution).
Dan says that he's always looking for new CBR applications. Please get in touch with him if you think of an intriguing possibility.
Calton Pu (email@example.com) was born in Taiwan and grew up in Brazil. (He still helps Brazilian expatriates keep in touch via email.) He received his Ph.D. in Computer Science from UWashington in 1986, then became an assistant professor in the Dept. of Computer Science at Columbia University. He is also a member of IEEE and ACM.
Calton is leading three research projects. The Harmony project has two aspects: 1) superdatabases that provide atomic transactions across heterogeneous database boundaries, and 2) high performance query/transaction processing in very large distributed database systems. A superdatabase called Supernova is operational. Calton recently introduced epsilon-serializability -- a generalization of transaction processing -- which increases database concurrency, availability, and autonomy.
His second project, also operational, is the Synthesis distributed operating system. It combines an orthogonal high- level kernel interface with efficient execution obtained from dynamic kernel code synthesis.
Calton's Netwatcher project measures end-to-end application performance over wide-area networks. He is also working on extended transaction models for long, open-ended activities, and on application of kernel-code synthesis to distributed software systems.
GTE Laboratories is looking for a LISP-cognizant summer student. Their DAI project is developing a multiagent architecture for customer network control and cooperative operation support systems. They want help building domain models using CMU's PRODIGY planning system. The job also involves research on an interagent language and on multiagent joint planning and coord- ination strategies. Contact Robert Weihmayer (firstname.lastname@example.org), (617) 466-2811, by the end of May for details. [DAI-List.]
The Institute for Supercomputing Research (ISR) in Tokyo is seeking a neural-network researcher for a one- or two-year visit. The work will involve basic and applied research in character recognition and optimization. Other basic research at ISR includes supercomputer performance evaluation, architecture, visualization, neural networks, multimedia information systems, and network/system management. For further information, contact Shaun Lawson, email@example.com, 011-81-3-536-7769 Fax. [Neuron Digest.]
The Bellcore Workshop on High-Performance Information Filtering: Foundations, Architectures, and Applications, will meet in Chester, NJ, this November 5-7. The workshop will examine filter construction, usage, and evaluation for information domains such as news, entertainment, advertising, and community information. Topics include: taxonomy of information domains and their dynamics; information retrieval and indexing; information delivery architectures; adaptive and distributed filtering agents; and cognitive models. By invitation only. For details contact Shoshana Hardt-Kornacki (firstname.lastname@example.org). [Connectionists.]
A similar event is the 2nd International BANKAI Workshop, to be held by The Society for Worldwide Interbank Financial Telecommunication (S.W.I.F.T.) in Brussels, October 14-16. Topics in intelligent information access include hyperlinked online manuals; NL, multimedia, and groupware databases; and data filtering/display for real-time data feeds. Contact B. Diraison, AI Lab, S.W.I.F.T., Avenue Adele, 1-B 1310 LA HULPE (Brussels?).
The May issue of CACM has a nice essay by Peter Denning. He denigrates the view of scientific research as input to human knowledge, with teaching as an output function. Instead, he espouses research and publication as a process of convincing the community that a new fact is true and usable, with teaching being the training of others to participate effectively in this process. As a corollary, Denning sees little merit in measuring scientific contribution by publications and citations. Impact on society can be through "seminars, symposia, professional meetings, laboratory visits, ..., writing articles, press releases, the electronic networks, and more." And through training of students. (If you take that view, though, don't forget to convince your tenure committee.)
Calton Pu (email@example.com) and John Josephson (firstname.lastname@example.org) have asked me about courting NSF grants. I appreciate the prompt. The following is personal opinion, and does not represent official policy of NSF. It is based on my experience in CISE/IRIS, and may not apply to other divisions -- especially those using panel reviews.
First, you'd better hurry if you are interested in this year's Small Business (SBIR) competition. The deadline is June 17. For literature, contact Roland Tibbetts ((202) 357-7527) or any NSF officer.
For other proposals, it's still a bit early to submit. End- of-fiscal-year proposals get the quickest evaluation and response, but are likely to be declined simply because no money is left. A proposal submitted in May could be held into the next fiscal year, but is more likely to be cleared out with the proposals that arrived in November and December. You would be competing with essentially the entire year's submissions, if there is money remaining at all. (A bit of money usually turns up at the end of the year, but the program director (PD) will have kept several of the better proposals available just in case. This is the group you must beat.) If no money remains, you will be declined unless at least half of your reviews are marked Excellent and the rest are enthusiastic enough to merit funding from the next year's money. New PDs are likely to hold over several such proposals, but experienced ones know that robbing the next year's budget just creates worse problems later on. The higher-ups also frown on proposals being held over, although they don't monitor closely.
It takes about a month for new proposals to be logged in -- one of life's little mysteries -- so a June proposal won't hit the PD's desk until July. That leaves only one month before fiscal- year action is closed out in the first week of August, so there really isn't time to get reviews back. Your proposal will be held into the next fiscal year, probably for the full length of time permitted. NSF is promising action in nine months, but declinations of moderately good proposals sometimes take longer than that. (They'll give you a quick declination if you insist, but for an award they want to see what other proposals are competing for the funds.) Be happy if you get money by the start of the next summer.
There's a slot on the cover page where you can specify when you want to start work. It is routinely ignored. If you really do need an early decision, say for a January start, say so in your proposal and in the transmittal letter. Unfortunately, transmittal letters are filed at the back of the proposal jacket and are often overlooked. A sticky-note on the front of the PD's proposal copy may be more efficacious. Contacting the PD every month or two is also likely to work. (Squeaky wheels do get action. Most PDs are professional enough that their annoyance with your calls will not affect funding decisions.)
The bulk of the proposals arrive in November. The target date keeps moving up as NSF's workload grows, but it really doesn't matter. It takes months to assign reviewers, so proposals arriving in December are at no real disadvantage. January and February proposals will be held until after PYI and RIA panels, so they do suffer. Really great November/December proposals might be funded by the start of summer, but others will have to wait until at least August.
My advice is to beat the rush. Some PDs spend out very early, so June-August proposals have the best chance. Others hold onto their funds as long as possible, so early proposals are at a disadvantage. (How can you find out? Try asking the PD! If he or she is too inexperienced to know, assume that early proposals are favored.) September/October proposals apply a bit of pressure for an early decision, but are late enough that the PD has a good feel for the competition before the decision must be made. The earlier the proposal, the more likely it is to get a careful reading instead of summary dismissal for lack of funds.
If your proposal is really great, though, submit it as soon as possible -- even in March/April/May. It might win quick funding, and the worst that can happen is that you get some feedback on how to improve it for resubmission by November. I suppose the same reasoning holds for not-so-great proposals. Roughly 25% of the proposals are funded (usually at less than was requested), so it makes sense to submit as many proposals as you can. (NSF claims more like 30%, but that includes special programs. In the regular research competition, I was having trouble reaching 20%. It gets worse every year.)
Suppose you've been turned down several times. Does that make it harder to win? I've seen it cut both ways. Sometimes a proposal comes in from a real flake, and a long series of past rejections can be taken as evidence that this researcher will never amount to anything. Other times a proposal is tempting, and a check of past declinations shows that they were also borderline. This can be taken as a sign that a little encouragement from NSF could have a big impact on the investigator's career. I remember one case where the PI had tried repeatedly to please various program officers, always falling just short. I felt it was time we quit jerking him around and let him get his research started. It was a pretty good proposal, of course.
Will a PD tell you if you're wasting your time? No. Not usually, and not directly. PDs are bureaucrats, and their first responsibility is to avoid gaffes that could percolate higher into the pyramid. Besides, your next proposal could be a winner -- it is never in the national interest to discourage submission of proposals. But if you sense any lack of enthusiasm for your work, it's time to get someone else to critique your proposals before you submit them. You should also have the PD send you program literature so that you can write your proposals around the currently requested buzzwords. Printed literature may be out of date, so ask whether the PD has any internal program documents that could be copied. (PDs have to prepare slides and statements at least once a year, and are often quite eager to distribute these.)
Should you visit? I suppose it depends on the PD, but I found visits to be an annoyance. PDs are exceedingly busy, and any visit of more than 15 minutes is disruptive. Presentation of past work is largely irrelevant, and future work will be evaluated by outside reviewers. Like most bureaucrats, a PD will be friendly but really has nothing to gain from your visit. Sales pitches by department heads and deans are the worst, since the PD doesn't even get to meet the investigator. And videotapes are like salesmen that can't be cut short.
The main purpose of a visit should be to target your proposal to a specific pot of money. Gather all the literature you can so that you can customize your proposal. (Besides, handing out literature gives the PD a feeling of accomplishment.) But all of that can be done by phone and by mail. If you know enough to make an appointment with the right PD, you don't need the visit.
There is seldom any bar to submitting multiple proposals. Each should be significantly different from the others, of course. A difference in scale (e.g., number of graduate students employed) is not considered sufficient, since that's negotiable in any case. A difference in viewpoint or sales pitch is also not sufficient, unless the submissions are in different special competitions. (Many people compete in both RIA and PYI, for instance.) But a real difference in scope of the work does open the way for multiple submissions, either within one program or across several. (You have to call this to NSF's attention in either case, so there isn't any real difference.) Be aware, though, that an award recommendation on any one of the proposals -- no matter how small -- may block you from winning on the other(s). PDs tend to view any award as your "share" for the year, and may even ask you to withdraw competing proposals.
I've been surprised at the willingness of investigators to withdraw their proposals. Perhaps the willingness is at the dean's level. Withdrawals are recorded right along with awards and declinations, so there isn't much effect on the investigator's record. Now, there are times when a proposal must be withdrawn or rejected as inappropriate for review. If you win a PYI award, for instance, you are no longer eligible for the RIA competition. And certainly the government cannot pay you twice for the same work. In general, though, a request for a withdrawal is for the PD's convenience rather than for your own benefit. It saves the PD from having to write an awkward rejection letter, or from sending a proposal for review when it hasn't a chance of being funded. The latter is justified, of course, but denies you the peer review and feedback that you would have gotten.
NSF does have a policy of awarding no more than two summer months of funding to any investigator. The assumption is that any professor has nine months of academic-year support and one month of vacation, so there are really only two months left. If that time is already covered by grants, even non-NSF grants, NSF will refuse additional support. Unless NSF makes an exception. Or doesn't notice that there's a problem. (It happens.)
There is no such limit on academic-year awards. Non-NSF academic-year awards are discounted as being none of NSF's business -- they don't affect NSF policies on summer funding. And NSF is free to award its own academic-year support. Some PDs refuse to do it under any circumstances; others will do it only for postdocs or researchers on soft money. The point is that you can ask; you just aren't likely to get it. Such budget negotia- tions rarely influence the likelihood of winning an award -- unless your request is so outrageous that reviewers comment upon it.
The gray area is that third summer month. Suppose you already have August paid for -- will NSF give you June and July? Probably. Will they give you July and August if June is covered? Probably not. It all depends on the specific program director, and on how much money remains. It also depends on the time of year that the work starts, since a mid-summer award would cover July of the first year and June of the next. It also depends on how strongly you state your need for the funding. If you insist that you can't do the work in less time, an NSF bureaucrat is unlikely to go on record saying that you're lying.
On the other hand, take-it-or-leave-it offers are best taken. Sometimes the PD wants to give you more, but simply can't. In such cases, try to negotiate for additional funds in the next fiscal year. You can get a supplement for the first year of work plus the increased level of support in all succeeding years.