New graduate students in computer science often have a difficult time
choosing among research projects and may never develop a broad view of
the field nor think about the relative significance of efforts in
various directions. When new graduate students ask me for advice I
tell them to start by pretending that they are the lab director for
computer science at a brand-new research university and to come up with
a plan for how they'd populate their lab with projects. This is sort of
like the fantasy baseball leagues that are popular with kids, hence the
page title. In preparing a fantasy research lab plan, a student will
need to familiarize him or herself with a broad range of problems and
the current state of the art in solutions. This ought to give the
student more perspective in planning a career.
Students aren't sure what this should look like and therefore I've
prepared a sample fantasy computer science research lab agenda below.
Labus Novus, a.k.a. Philip's Fantasy Computer Lab
Research at Labus Novus is shaped around applications that we're trying
to build. Computer Science is an engineering discipline and therefore
we make sincere and serious efforts to ensure that our work will meet
human needs. We pick applications that are sufficiently beyond the
current state of the art that we can populate our laboratory with a mix
of medium-term, long-term, and blue sky (50+-year) projects.
Americans make expensive employees. Productivity is measured as
economic output per dollar of labor input. In the absence of
technological advances, the only way to improve labor productivity is to
move the job to a low-wage country. Pairing every American office
worker with an assistant in a low-wage English-speaking country (or a
low-wage part of the US) would be an excellent way of boosting
productivity without exporting jobs, assuming that an effective
coordination system can be constructed.
The Labus Novus coordination system will comprise the following
Current state-of-the-art video conferencing systems require 6 Mbits of
point-to-point bandwidth. Thus the requirement of extremely high
quality video conferencing implies the need for research in video
analysis and compression, network protocols and routing, and
semiconductors and optics for very bright images.
- a life-sized two-way video conferencing system; it is as if a wall in
one's office is opened up to the assistant's office thousands of miles away
- an information system that records everything relevant to the
high-wage worker's job, including facts, reference material, contacts,
correspondence, appointments, and relationships among these items [like
a more sophisticated Microsoft Outlook]
- robot arms and other robots within the high-wage worker's office
that can be manipulated by the low-wage worker, thus enabling the
assistant to pull folders from file cabinets, position papers on a desk,
etc., from the other side of the planet
Outlook-on-steroids might sound straightforward but doing the job right
can be as challenging as all of Artificial Intelligence. We are
building support for a computer-mediated assistant rather than
attempting to build a fully automated personal cognitive assistant.
This does not reduce the difficulty of achieving a complete solution but
it does increase the utility of an incomplete solution.
A desire to give the assistant the ability to manipulate physical
objects half a world away (telepresence) justifies research in broad
areas of robotic actuators and sensors.
Funding Possibilities: Phone companies are logical sponsors for this
research. Telcos built a tremendous amount of network capacity in the
1990s but then neglected to offer any services besides voice
communication, thus resulting in falling prices and bankruptcies. Only
about 10 percent of the fiber installed through the U.S. is actually
being used. Continuously active high quality video conferences have the
potential to consume all of currently unused bandwidth in the networks.
Note that the system could be used domestically, yoking together a
worker in an expensive crowded place such as New York City with an
assistant in a low-wage uncrowded place such as Iowa.
Wandering down to the self-help section of bookstore one discovers a
world of unmet human goals. A lot of these goals are tough to reach
because we lack willpower. Olympic athletes also lack willpower at
times, yet they get to the Olympics and we're still fat. Why? Maybe
because they have a coach and we don't. You might argue that society
would be better off if only half of us worked and lived and the other
half devoted themselves to coaching but it is unlikely that 50 percent
of our citizens would want to give up all of their aspirations. Thus
Labus Novus will build a network-resident computer-based personal coach.
Our coach needs to be able to keep track of our goals and what we've
accomplished toward those goals. For example, if our goal is losing
weight the coach needs to know our most recent weight and about any food
intake or exercise. The recency requirement means that we need to be
able to communicate with our coach regardless of where we are, by
telephone, by handheld computer, by desktop computer, or entirely
passively (computer notices that we've stepped on our home scale and
records the weight).
Ensuring coach availability implies research on ubiquitous wireless
Internet connectivity. Making a coach that isn't too cumbersome
requires a conversational speech interface, thus implying research at
the very edge of current speech systems. Building a coach that is
unobtrusive will require cameras mounted in homes and work places and
machine vision so that the coach can figure out what you're doing (no
sense interrupting you if you're in a conversation but if you're alone
and reaching for the cookie jar it might be time to admonish). A world
full of cameras requires research in computer and network security so
that only your coach has access to your private life. The coach needs
to be smart enough to learn something of your habits and thus implies
research in machine learning.
If we consider the domain of health and weight coaching it is useful to
build wearable low-power hospital-grade instrumentation for all aspects
of the human body, thus driving research into miniature electronics and
sensors. Anything that can be measured should be measured and recorded,
automatically and without user intervention. At the same time we can
interface computer graphics and games to exercise equipment, to motivate
the coachees, thus opening the door to all of the graphics research that
goes into video games (but this time aimed at making people less fat
rather than more).
Funding Possibilities: the military spends a tremendous amount of money
on training and physical conditioning; they ought to be willing to
invest in a technology that would help each soldier be all that he (or
she) can be.
It is tough to know whether or not one's ideas in high-performance
computing are right but it is easy to tell whether or not it is raining
outside. Thus Labus Novus chooses to push the state of the art in
parallel processing, scientific computing, and compilers by trying to
predict the weather.
People seem to be happy with computers that sometimes crash but unhappy
with airplanes that sometimes crash. Thus we focus our research in
software specification, proving programs correct, operating systems, and
constructing fault tolerant hardware on the application of avionics.
Labus Novus will build its own operating system from the ground up.
Academicians keep saying that safe high-level languages such as Haskell,
Lisp/Scheme, ML, etc. are more reliable and result in higher programmer
productivity than unsafe low-level languages such as C. However in
practice nearly all computer science research is done on top of standard
old-style operating systems such as Unix and Windows. If high-level
languages are so great, it ought to be possible for a university or a
consortium of universities to build something better and more reliable
than Unix in short order.
Simulated flight is safer and more practical in the frigid Northeast
than actual flight, especially for uncertificated students. Thus our
goal of better avionics implies a heavy workload for the graphics
researchers in our lab.
Throughout an airplane are opportunities for microelectromechanical
systems (MEMS), starting with the gyros used to sense airplane attitude,
velocity, and acceleration.
An airplane is a acoustically noisy environment. Designing active
noise-cancelling systems for airplanes can take a researcher into a wide
range of computational acoustics, microelectronics, and real-time
Avionics require high-reliability high-availability databases of terrain
information, instrument flight approach procedures, airspace
restrictions, and airport information. This information is constantly
changing and it ideally would be updated from intermittent radio contact
with the Federal Aviation Administration's extensive transmitter
network, thus entailing research on distributed database management.
Pilots are challenged with a flood of constantly changing information,
thus opening the door to a lot of user interface and human-computer
Funding Possibilities: the Air Force.
Modern codes are very difficult to break with silicon-based computers.
A biological computer offers the possibility of dumping a lot of goop
into a beaker, each cell of which is effectively following one branch of
the tree of possible solutions, and coming back a day later to see if
one of those cells happened upon the answer.
Labus Novus as sketched has people working on the following areas:
This is more or less the full spectrum of activities going on at the top
computer science labs and yet the mission and value of the Labus Novus
can be clearly explained to a layperson in two minutes. What's more the
progress of the lab can be measured easily by combining intermediate
solutions into test applications and seeing if they work well.
- speech recognition
- machine learning
- video processing, representation, and compression
- video projection
- machine vision
- high-speed wired networking
- wireless networks
- languages and compilers
- operating systems
- theory (proving programs correct)
- fault-tolerant hardware design
- parallel processing and high-performance computer architecture
- distributed database management systems
- user interface and HCI
- bio-electrical interfaces
- computational biology
Here are some things that I'd toss in because I like them. Some are
near-term and straightforward and would make good projects for master's
Software Support for Photographers
Commercial tools such as Adobe PhotoShop address the needs of graphic
artists fitting a single photo into a spot on a printed page. Labus
Novus will build a system that directly supports photographers'
aesthetic goals. First and foremost is assisting with the challenge of
mapping the real world's very high contrast into the narrow range of
contrast that can be displayed on a monitor or printed on paper. Second
we would like to give photographers aesthetic and stylistic controls
rather than options such as "Gaussian blur". Third we want to allow the
processing of photos in a group, e.g., "let's see whether my 85 vacation
snapshots from Greece will look better in black and white." Fourth we
want to support the presentation of photos on the Web and to mobile
Telephone Solicitor Interlocutor
To a master's student looking to do something with speech, I'd say
"build a system that will occupy telephone solicitors." The challenge
for the computer is to keep the phone solicitor under the impression
that he or she has reached a human being for as long as possible. The
beauty of this system is that if installed in a wealthy area there would
be a near-endless stream of calls on which to test the quality of the
result and that, if a high quality system were widely installed on PCs
nationwide, it would put the phone solicitors out of business (because
they'd be spending so much time and money talking to computers
programmed to keep them on the line indefinitely).
Companion: National multi-school contest for the best system.
This is the end of the sample fantasy lab writeup.
One of the things that players of fantasy sports enjoy is competition,
i.e., the ability for a computer to simulate Fantasy Team A playing
Fantasy Team B. At first glance it would seem that Fantasy Computer
Science Research Lab is not amenable to competition, thus depriving
students of the chance to excel and win. Yet many schools have business
plan competitions, judged by human experts. MIT's is called the "50K"
and includes a $50,000 cash prize. If a university can have a contest
for short-term business ideas why can't it run a contest for long-term
research ideas? In fact we could have a nationwide contest for first-
and second-year computer science grad students (only about 1000 graduate
each year with PhDs so there aren't an overwhelming number of these
folks). Each student would be limited to 4 pages. Judging would be
done by people from funding agencies such as NSF and DARPA or funders of
university research from industry.
The grand prize? One year of funding, a post-doc to help with the
research, and a travel budget so that the student could go to
conferences and tell people about what was accomplished.
An alternative: Reality Computer Science Research Lab
For those who prefer to deal in brutal truths than in fantasy, the best
place to start planning a lab is probably www.darpa.mil. If the generals don't
want to fund it, you probably aren't going to be able to do it. The
other major government sponsor of computer science research is
www.nsf.gov and they are fairly
explicit about what they want as well.
Text and photographs copyright 1990-2003 Philip
I think the idea of a clearly explainable goal, or focus, of the lab is critical. I the idea that all the technical projects can fit under one umbrella that even a five year old could understand.
Add a comment | Add a link
-- Jacob Jans, May 14, 2003