Volume 10, Number 2
                      condu t          Department of Computer Science                                   Fall, 2001
                                                    Brown University

                                                                        I wondered whether this was an artistic de-
                                                                        cision or whether it contributed to the dy-
ADAPTING HUMAN PERFORMANCE                                              namics of the walking motion. According to
   FOR HUMANOID ROBOTS                                                  one of the long-time researchers on this
                                                                        project, the dynamics of the arm swing ac-
                                                                        tually are important, and swinging the
                                         How can we make                arms in this way saves them about 16% of
                                         humanoid robots                the total energy required for the walk. This
                                         expressive, useful,
                                                                        is good news for those of us who believe
                                         or entertaining? Is
                                                                        that concepts such as energy minimization
                                         there a market for
                                                                        are important for understanding human
                                         humanoid robots?
                                                                        (and humanoid) motion.
                                         How can we make
                                         them safe to be                Sony has entered this area of humanoid re-
                                         around?                        search more recently. Their 50cm high
                                         I had the opportu-             Sony Dream Robot (SDR) has been posi-
                                         nity to work in Ja-            tioned as the successor to the Aibo dog.
                                         pan for a few                  With their focus on entertainment, Sony
                                         weeks this sum-                may be the first actually to turn a profit on
                                         mer, and human-                humanoids. We saw the SDR kick a ball in
                                         oid robots were a              a videotaped demo, so it is clear that Sony
                                         recurring theme of             is thinking about Robocup (www.rob-
                                         the trip. There is a  soccer tournaments with the SDR
                                         large    concentra-            as a competitor.
                                         tion of humanoid
 Nancy Pollard
                                         research in Japan,          “we heard about and
                 including significant corporate efforts by
   and DB        Honda and Sony. Honda has been working               saw humanoid robots
                 on humanoid robots for 15 years now.
                 Their interest in humanoids is quite un-
                                                                    that could fall down and
                 usual when you think about it, given their               get up again”
                 primary business of cars! You have to ad-
                 mire their ability to think long-term.                 There is also a great deal of humanoid re-
                 Honda is now expressing interest in mak-               search at Japanese labs and universities.
                 ing its humanoid research program begin                At a workshop with the theme “New Fron-
                 to pay for itself, however, and their child-           tiers in Intelligent Robotics,” hosted in To-
                 sized Asimo robot will be available for
                                                                        kyo by the Japan Society for the Promotion
                                                                        of Science and Professor Inoue from the
                 As an aside, here is an interesting fact               University of Tokyo, we heard about and
                 about Asimo. If you see it walk, it swings             saw humanoid robots that could fall down
                 its arms vigorously—more so than the                   and get up again, humanoid robots whose
                 other Honda robots. When I first saw this,              motion was controlled by a network of ten-
                                            Brown University, Box 1910, Providence, RI 02912, USA
             dons instead of the usual motors, and “so-            oid may be the most compact general-pur-
             cial” robots that were programmed to                  pose robot possible.
             interact with and entertain children. In-
                                                                   (2) People may be better able to relate to
             formation on these and other humanoid
                                                                   robots that look something like ourselves
             efforts can be found at
                                                                   and that can communicate in a similar
             Along with a few friends and colleagues—                  (3) It should be easier to teach robots that
             Jessica Hodgins and Chris Atkeson from                    are similar to ourselves. Ideally, we would
             CMU and Marcia Riley from Georgia                         like to be able to just show the robot how
             Tech—I spent a couple of weeks in Kyoto                   to do something and have it perform the
             working with DB, yet another humanoid                     task correctly.
             robot, but more about that shortly.
                                                                       (4) There is expected to be a growing need
             WHY HUMANOIDS?                                            for robots in the home. The Japanese pop-
                                                                       ulation is aging very rapidly compared to
             One obvious question to ask is, “Why all
                                                                       other nations, with 25% of the population
             this interest in humanoid robots?” Here
                                                                       expected to be over age 65 by the year
             are some of the reasons I heard during
                                                                       2020. There is a great deal of interest in
             my visit:
                                                                       service robotics, especially robots for the
             (1) Man-made artifacts in the world have                  home, and many imagine that these ro-
             largely been designed with people in                      bots will be at least partially humanoid in
             mind. A humanoid robot should be able to                  appearance.
             go wherever we go, reach whatever we                      (5) The basic desire to reproduce our-
             can reach, use whatever tools we can use.                 selves probably figures in at some level
             Some researchers believe that a human-                    as an additional motivating factor!
                                                                   THE HUMANOID ROBOT DB
                                                                   Humanoid robotics research seems to fall
                                                                   into two main camps at the moment that
                                                                   can loosely be described as navigation
                                                                   and upper-body skills. A large proportion
                                                                   of humanoid robotics research focuses on
                                                                   the problem of transport—getting the ro-
                                                                   bot simply to traverse hard floors, car-
                                                                   peted floors, and sloped floors and move
                                                                   up and down stairs is difficult, and biped
                                                                   robots cannot yet walk on uneven terrain.
                                                                   A completely separate set of problems has
                                                                   to do with giving the robot the ability to
                                                                   manipulate objects and communicate in
                                                                   an expressive manner. The two sets of
                                                                   problems are often decoupled: if naviga-
                                                                   tion skills are the primary research focus,
                                                                   manipulation skills of the robot may be
                                                                   limited; if manipulation or expressiveness
                                                                   is the primary research focus, the robot
                                                                   may have wheels or be fixed in place.
                                                                   The robot DB was constructed to study
                                                                   higher-level functions of the brain, such
                                                                   as learning from demonstration. DB has
                                                                   30 controlled degrees of freedom, but true
                                                                   to form, DB cannot walk. Although its
                                                                   legs do move, it is rigidly mounted in
                                                                   place by a mechanical connection at the
The humanoid robot DB in action. The paper towels                  pelvis. Skills that have been explored us-
under his feet are to soak up leaking hydraulic fluid!             ing DB range from drumming to visual

                                                          conduit! 2
     tracking to Okinawan folk dance in re-                 is mapped to a skeleton used for anima-
     search supported by the Kawato Dy-                     tion purposes. This process also intro-
     namic Brain Project at ATR in Kyoto, Ja-               duces error, because the joints of the
     pan (                 skeleton may not exactly match the joint
     me.html).                                              locations of the performer, and the skele-
                                                            ton does not allow the same types of mo-
     EXPRESSIVE ROBOTS?                                     tions as a person. It cannot, for example,
     Our interest in DB is to try to understand             capture the full flexibility of the human
     how human performances can be adapted                  spine.
     to the robot with as little quality loss as
                                                            w Restricting the motion to the degrees of
     possible. We chose entertainment—story-
                                                            freedom of the robot. The robot has fewer
     telling—as a domain because this robot is
                                                            degrees of freedom than the animation
     well-suited to the task and because story-
                                                            skeleton. The most substantial difference
     telling utilizes a wide variety of motions
                                                            is that the robot has a completely rigid
     exercising a large portion of the per-
     former’s workspace.
     It is not straightforward to map a human               w Restricting the motion to the robot’s
     storytelling performance to DB because                 joint limits. The robot has a much smaller
     the robot has many fewer degrees of free-              range of motion than the human subjects.
     dom than our human subjects, different                 One perceptually salient example is the
     limb lengths, a different range of motion,             fact that the elbow joint cannot be fully
     and different velocity limits. In fact, we             extended.
     can think of the process of transferring
                                                            w Restricting the motion to the robot’s ve-
     motion from the human actor to the robot
                                                            locity limits. The robot also has somewhat
     as putting the motion through a series of
                                                            restrictive velocity limits. Velocities are
     filters, each of which introduces error.
                                                            limited when the robot nears the maxi-
     These filters are:                                      mum torque that can be supplied by the
     w Capturing the motion. We used an opti-
     cal motion-capture setup in Jessica Hod-               w Controlling the robot to track a desired
     gins’ lab at CMU. Eight cameras record                 trajectory. The robot will not go exactly
     motion of reflecting balls placed on our                where it is commanded to go, which leads
     subjects as they perform. The main                     to another source of error.
     source of error at this step is motion of
     the reflecting balls with respect to the
     skeleton of the performer due to motion of
     the clothing or skin.

     w Mapping the motion to an animation
     skeleton. After the motion is captured, it

                                                                Our animation skeleton has joints
Jessica Hodgins’ motion-capture lab at CMU                           in the locations shown

                                                   conduit! 3
       Our first pass, which we implemented                    mance by each level of filtering? We are
       while in Japan, was to obtain as similar a             designing a user study to be run this fall
       “rendering” of the original motion as pos-             that will test subjects’ ability to match
       sible. This meant throwing out degrees of              motion samples with videos of the origi-
       freedom that the robot does not have, lo-              nal actor performances. (Was this motion
       cally compressing and smearing the mo-                 obtained from this actor?) We expect that
       tion to fit within joint and velocity limits            the task will be trivial when the motion
       respectively, and adding a learning phase              sample is taken from raw motion-capture
       so that the robot would perform the re-                data, but that it will become difficult
       quired trajectory as accurately as possi-              when we present motion that has passed
       ble. The figure shows a snapshot of some                through a number of filters. In particular,
       of our results in a single frame of the ro-            the filter for the joint limits of the robot
       bot and human performers doing the chil-               often results in a substantial change to
       dren’s song “I’m a Little Teapot.”                     the motion. It is also possible that the me-
                                                              chanical appearance of the robot will be
       How do we evaluate these results? In                   sufficiently distracting to make the task
       other words, how do we assess the extent               more difficult than if the motion were
       of the damage done to the original perfor-             played through an animated humanlike

                                                              Even before the user study results are
                                                              available, we can see room for improve-
                                                              ment in the robot’s performance. One flaw
                                                              is that we treat each degree of freedom
                                                              (each motor) separately, scaling the mo-
                                                              tion of that degree of freedom into the
                                                              joint and velocity limits of the robot.
                                                              What a person observes, however, is the
                                                              motion as a whole, and meaningful ges-
                                                              tures that comprise that motion. These
                                                              gestures are a result of the coordinated
                                                              efforts of many joints, and so these joints
                                                              cannot be treated independently. For ex-
                                                              ample, the robot has a very restricted
                                                              range of motion side to side (abduction/
                                                              adduction) at the waist and a more gener-
                                                              ous range of motion forward and back
                                                              (flexion/extension). This difference in lim-
                                                              its, along with our relatively simple pro-
                                                              cessing technique, results in the following
                                                              artifact: when the robot bends to the tune
                                                              of “tip me up and pour me out,” it first
                                                              leans mostly to the left or right, hits the
                                                              abduction/adduction joint limit, and then
                                                              continues to lean forward, making the
                                                              tipping motion look more circular than
                                                              linear. This sort of problem is easily
                                                              solved if we have a model of meaningful
                                                              gestures. The tipping motion could be
                                                              mapped in its entirety into a linear mo-
                                                              tion within the robot’s workspace.

                                                              Even though there is room for improve-
                                                              ment in the robot’s performances, I
                                                              strongly believe in the importance of hu-
                                                              man examples for creating realistic mo-
                                                              tion of both animated characters and
DB trying to follow a human performance of                    robots. On an orthogonal track, I am ex-
             “I’m a Little Teapot”                            ploring how to design a robot hand that is
                                                     conduit! 4
              mechanically practical and has similar                  an anatomically motivated hand design.
              force and stiffness capabilities to the hu-             But perhaps this is the topic of a future
              man hand. I hope to show that designs                   article.
              that are similar to the human hand in
              these ways will facilitate teaching grasp-              FINAL THOUGHTS
              ing and manipulation skills by example.
              In other words, it should be easier to                  This project was in part an excuse to visit
              grasp and manipulate objects successfully               Japan and spend some time working at
                                                                      ATR, but it did raise some very interest-
                                                                      ing questions that we look forward to ad-
                                                                      dressing in future research, such as what
                                                                      characteristics or features of motion
                                                                      should be preserved in a mapping from
                                                                      human actor to animated character or ro-
                                                                      bot. I also enjoyed the overview of the cur-
                                                                      rent state of the art in humanoid robotics
                                                                      research. There are of course many issues
                                                                      to be resolved before we have C3PO wan-
                                                                      dering around, not the least of which is
                                                                      safety of large autonomous devices that
                                                                      are about as stable as inverted pendu-
                                                                      lums. Humanoid robotics research has a
                                                                      great deal of support behind it right now,
                                                                      however, and with the added ability eas-
Jessica, DB   following human examples with such a                    ily to capture large datasets of motion ex-
and Nancy     hand than with existing hand designs                    amples, we can expect to see humanoid
              that are not so strongly based on human                 robots becoming more graceful, expres-
              anatomy, simply because the passive dy-                 sive, and perhaps somewhat useful in the
              namics of the hand work in your favor in                near future.

              Mark Dieterich, senior systems administrator for the graphics group, is also an amateur ra-
              dio operator. When disaster strikes, the Federal Government activates the Radio Amateur Civil-
              ian Emergency Services (RACES) and places its operators on standby. After the WTC tragedy,
              New York City’s operators were immediately called into action to provide non-secure radio net-
              works for emergency personnel working at ground zero.
              It soon became necessary to introduce fresh operators from surrounding areas and Rhode Island
              operators were put on standby. When the call came the weekend after the attack, Mark and four
              others took their equipment to the Red Cross’s initial staging area in Westchester County. From
              there they were bused to the Red Cross in Brooklyn and thence to the site in lower Manhattan.
              Cell phones in the area were next to useless and a stable communications network was critical
              for smoothly coordinated rescue efforts. The amateurs operated on a single frequency with a con-
              trolled flow of information and requests going to a net controller and then out to other operators
              assigned to various tasks. Mark was responsible for directing calls to various Red-Cross-run
              shelters: respite centers for police, firemen and other workers, shelters for dislocated people and
              secure government shelters. He also spent a day ‘tailing’ a Red Cross executive around the site.
                                     For the last four days of his week-long stay, Mark became the day-shift
                                     manager for amateur radio operators. His job was to brief and debrief
                                     teams going into and out of the field and to deal with non-routine prob-
                                     lems. Most 12-hour shifts were in fact 18 hours long, as they still continue
                                     to be.
                                     Mark was a dispatcher for his hometown (Pittsford, NY) volunteer ambu-
                                     lance company for over four years, so he has experience with emergency
                                     situations; however, he found working at the WTC site overpowering. As
                                     the week progressed, the tight security became increasingly evident, with
                                     MPs on guard outside Red Cross headquarters. Said Mark, “If you knew
                                     New York before, it’s changed—the atmosphere—people are making eye
                                     contact in the subway and are talking to one another instead of reading.”
                                     Anyone involved with rescue and cleanup has found restaurant owners re-
                                     fusing to accept payment, taxi drivers who turn off their meters when a
                                     worker gets in; for workers, the subway is free. Being of use has helped
                                     him cope with the enormity of the tragedy.

                                                             conduit! 5
                                       SPINNING THE SEMANTIC WEB
                      The folks from conduit! contacted me a                  demic AI products ever—but it would still
                      while back and asked if I would consider                be being used by fewer than one in every
                      writing a short article for them. Having                100,000 web users! To really change the
                      become a conduit! fan and being happy                   web, and as you’ll soon see that’s my goal,
                      to repay the CS Department and faculty                  something bigger was needed.
                      for all the time and energy they put into
                                                                              Unfortunately, to do something on a
                      educating me so many years ago, I agreed.
                                                                              larger scale there’s not a lot of places to
                                           At the time, I didn’t              play. Academics wanting this sort of im-
                                           realize the challenge              pact have generally had to choose be-
                                           of combining a dis-                tween two alternatives: start your own
                                           cussion of my work                 company or join some large software con-
                                           with some biogra-                  cern (generally one located in Redmond,
                                           phy—but the two are                WA). Neither of these appealed to me, as
                                           intertwined and the                corporate America has never been an
                                           editor asked for both,             arena I found attractive. But I found a
                                           so here’s my best                  third alternative—if you want to work at
                                           shot.                              web scale, join the folks who made the
                                          Let me start with a                 web happen!
                                          little biography. Up-               The Internet, as we now call it, started
                                          on completing my                    out as a project called the “ARPAnet” un-
                                          doctorate at Brown                  der the sponsorship of the Defense Ad-
                                          in 1986, I joined the               vanced     Research     Projects    Agency
                                          faculty at the Univer-              (DARPA1) of the U.S. government.
                                          sity of Maryland and                DARPA also helped to support work in
Jim Hendler, PhD ’86 started a research group in artificial intel-
                                                                              the creation of the World Wide Web and
                     ligence. With a few lucky breaks and a lot               in tools for using web information. Fur-
                     of great students, I managed to get pro-                 ther, DARPA is currently the govern-
                     moted a couple of times and became a full                ment’s largest funder of research in
                     professor with a large research group.                   computer science. It was clear that to
                     Much of my group’s work centered on                      have the sort of impact I wanted, DARPA
                     scaling up work started at Brown and try-                was a place to consider. So in late 1998, I
                     ing to show that AI could be a valuable                  took a three-year leave of absence from
                     player on the then-emerging World Wide                   UMD and signed up to work at DARPA.
                     Web. A language we developed, called
                     SHOE for “Simple HTML Ontology Ex-                       Now I’d like to introduce my partner in
                     tensions,” gained a certain cachet in a                  crime in our effort to change the world. In
                     particular research community, and we                    the late 1980s, a researcher named Tim
                     duly published various boring academic                   Berners-Lee created a program called
                     papers, developed a number of small and                  World Wide Web that caught on pretty
                     inconsequential demos, and did all the                   well. With backing from various research
                     other things needed for me to get pro-
                     moted and my students to find academic                    1. Why ARPAnet and not DARPAnet? Gener-
                     jobs.                                                    ally, under Republican administrations or
                                                                              Congresses the organization has been asked to
                      Somewhere about four years ago, how-                    focus on the science needs of the DoD and has
                      ever, I realized that although the work                 been called DARPA. When the administration
                      was going well, it was also going rela-                 is Democratic, the D is dropped, and the
                      tively nowhere. The World Wide Web is an                agency becomes ARPA and focuses on dual-
                      unimaginably large cyberspace, and even                 use technology. The name changes often
                      the best academic efforts rarely get any                enough that the official coffee mugs say
                      real visibility on it. I thought that if I              “DARPA” on one side and “ARPA” on the other,
                                                                              so that employees won’t have to remember
                      could get, say, 10,000 people using my
                                                                              which one to give visitors.
                      stuff, it would be one of the most used aca-

                                                                     conduit! 6
        agencies including DARPA, Tim moved to                   The vision of the semantic web is not that
        MIT in the early ’90s and in 1994 started                of a new web but of new languages and
        the World Wide Web Consortium (W3C)                      functionalities that extend current web
        to help define and standardize the lan-                   functionalities. On the Semantic Web, not
        guages on the web. The consortium now                    just web pages, but databases, programs,
        has a membership of over 500 companies,                  sensors and even household appliances
        including all the largest players in infor-              will be able to present data, multimedia,
        mation technology and web applications.                  and status information in ways that pow-
        Berners-Lee is the director of the W3C                   erful computing agents can use to search,
        and remains one of the leading thinkers                  filter and prepare information in new
        about the future of the web.                             ways. New markup languages that make
        One of my first acts at DARPA was to talk                 significantly more of the information on
        to Berners-Lee about an idea of his called               the web machine-readable power this vi-
        the “Semantic Web.” The work my group                    sion and will make possible a new genera-
        had been doing in SHOE was aimed at                      tion of technologies and toolkits.
        some of his semantic web ideas, and I                    The semantic web is expected to evolve
        thought some joint effort between DARPA                  from the existing web as these new lan-
        and the W3C could help move this tech-                   guages and tools find their way into the
        nology out of the laboratory and into com-               marketplace. Figure 1, based on a figure
        mon use. Our interaction led to two new
        creations—a DARPA program called the                     in a recent Nature article4 on the future
        “DARPA Agent Markup Language”                            of electronic publishing, shows the poten-
        (DAML) and a W3C activity called “Se-                    tial evolution of the semantic web. We are
        mantic Web Advanced Development.”                        currently seeing new web languages like
                                                                 the eXtensible Markup Language (XML)
        I recently joined with Berners-Lee and                   and the Resource Description Framework
        another colleague to produce a Scientific                 (RDF) that let users produce more “meta-
        American article2 describing aspects of                  data” about web resources. These meta-
        this new work—I summarize below some                     data can be as simple as saying who
        of that article. I also steal from a forth-              produced the document and when, or
        coming article, “The Semantic Web: A                     complex enough actually to replace the
        Network of Content for the Digital City,”                web page with machine-readable infor-
        coauthored with a 14-year-old whiz kid                   mation.
        named Aaron Swartz3.
                                                                 To understand the difference in using this
        2. T. Berners-Lee, J. Hendler and O. Lassila,
                                                                 emerging net, imagine going to a search
        The Semantic Web, Scientific American, May,               engine and typing the query, “How many
        2001. (                        train lines are there in Japan?” If you use
        0501issue/0501berners-lee.html)                          any of the popular search engines, you get
                                                                 many, many pages back (at this writing,
                                                                 between 122,000 and 99,000,000 answers
                                                                 were returned)—and few, if any, actually
                                                                 contain the answer to the query being
                                                                 asked. As a foundation for browsing the
                                                                 web and learning lots about Japanese
                                                                 transportation, this may be okay, but for
                                                                 answering the question—worthless!

                                                                 3. A. Swartz and J. Hendler, The Semantic
                                                                 Web: A Network of Content for the Digital
                                                                 City, Proc. 2d Annual Digital Cities Confer-
                                                                 ence, Kyoto, Japan, October, 2001. (http://
                                                                 4. T. Berners-Lee and J. Hendler, Publishing
                                                                 on the Semantic Web, Nature 410, 1023-1024
 Figure 1: The evolving semantic web—new                         (26 April 2001) (
web languages and toolsets for revolutionary                     ture/debates/e-access/Articles/berner-
    functionality on the World Wide Web.                         slee.html)

                                                        conduit! 7
             Why, however, can’t this answer be found?                    other, a complicated program could be
             Many resources on the web could be used.                     written to find these, identify them, and
             First, some of the many documents found                      count how many there are—but writing
             probably have this information, but cur-                     such a program is a massive undertaking,
             rent language-processing and search tech-                    far more effort than users would want to
             nologies are nowhere near good enough to                     make for a single query.
             find them. Second, web-accessible data-                       We can imagine a day when a query like
             bases and programs could provide the an-                     this could elicit a very different sort of re-
             swer, but word-based text matching is not                    sponse. For example, a semantic web
                                                                          query tool could give replies like:
 “Sure, it’s a long way from here to                                         ➡ says
                                                                             the number of train lines is over 5000.
 there—and there’s no guarantee                                             ➡ There is a database that can provide
we’ll make it—but the possibilities are                                     that number; please provide an autho-
                                                                            rization number.
 endless, and even if we don’t ever                                          ➡ There is a web service that can com-
achieve all of them, the journey will                                        pute that number; please provide 500
                                                                             yen for the answer.
 most certainly be its own reward”
                                                                             ➡ I can get you an approximate an-
             sufficient to pull them out. Third, since                        swer by search and filtering, but it will
             each train line in Japan makes its pres-                        take about 4.5 hours to compute.
             ence known on the web in one way or an-                      The goal of semantic web research is to
                                                                          develop the languages and tools to make
                                                                          answers like this possible. (Details on
                                                                          current research can be found in a num-
                                                                          ber of places; two of particular use are a
                                                                          general information site about the se-
                                                                          mantic web at http://www.semanticweb-
                                                                          .org and the site describing the ongoing
                                                                          DARPA work at
                                                                          Does this sound like a crazy science-fic-
                                                                          tion dream or a lot of hype? A decade ago,
                                                                          who would have believed a web of text,
                                                                          conveyed by computer, would change the
                                                                          way we live and work? This new vision
                                                                          unites old players such as Berners-Lee
                                                                          and DARPA with new visionaries such as
                                                                          young Aaron Swartz, who recently
                                                                          summed it up well. Speaking about the
                                                                          semantic web, he said, “Sure, it’s a long
                                                                          way from here to there—and there’s no
                                                                          guarantee we’ll make it—but the possibil-
                                                                          ities are endless, and even if we don’t ever
                                                                          achieve all of them, the journey will most
                                                                          certainly be its own reward.”
                                                                          As for me, I’m now back at the University
          Academic office aide Akina Cruz, looking elegant                 of Maryland where I’m Director, Semantic
          in traditional African dress brought back on a                  Web and Agent Technology at the newly
          recent trip to Ghana by her sister, Adeola Oredola,             formed Maryland Information and Net-
          a senior at Brown. Despite its exotic look, this out-
          fit is called simply a skirt and headwrap.                       work Dynamics Laboratory. This lab is an
                                                                          attempt to develop new interaction modes

                                                             conduit! 8
                  between academia and industry; we hope                  I look forward to hearing from Brown col-
                  to create an organization that can not                  leagues, old and new, interested in the se-
                  only perform research into new web and                  mantic web vision.
                  network ideas, but also help transition       
                  these ideas into use more quickly.

                                                                          also played basketball, enjoyed skate-
                                                                          boarding (yes, he watched some of the
    NEW CS FACULTY MEMBER                                                 Gravity Games), biking and bodybuilding.
                                                                          His favorite reading subject is popular
                              Ugur spent his first ten                     physics, followed by history as a close sec-
                              years in Ankara, Turkey, be-                ond. He’s also a classical rock fan (Pink
                              fore moving to Izmir on the                 Floyd was one of his early favorites, al-
                              west coast. Izmir is Turkey’s               though he couldn’t understand a word at
                              third largest city, very beau-              the time!). Ugur looks forward to getting
                              tiful and, according to Ugur,               together with Pascal Van Hentenryck, a
                              the best city in Turkey to                  one-time provincial-level soccer player in
                              live in. His father is a consti-            Belgium and senior faculty member, to
                              tutional judge (i.e. not in-                play soccer and perhaps start a CS team.
                              volved with litigation), his                Ugur’s wife, Gamze Tunali, is still work-
                              mother a bank manager, now                  ing in Maryland at a networks company.
                              retired, and his older brother              She hopes to join him in Providence be-
                              works in the insurance in-                  fore the end of the year. She and Ugur
                              dustry. He returned to An-                  were classmates at the Bilkent Univer-
                              kara to attend Bilkent                      sity; they were married in Maryland and
                              University as a CS major,                   remarried the following year in Turkey
                              when he was 19. Given the                   with both families present. Ugur spends
Ugur Cetintemel               stiff university entrance ex-               almost every weekend in Maryland with
                  aminations and intense competition for                  Gamze; consequently, he hasn’t even seen
                  university places in Turkey, Ugur’s full                a WaterFire yet, but is keen to do so.
                  scholarship award was a major accom-
                  plishment.                                              Living in Providence means everything is
                  After receiving his MS at Bilkent, Ugur                 within walking distance and easily acces-
                  came to the US for graduate work in CS                  sible—no more 45-minute commutes. He
                  at the University of Maryland, College                  enjoys the many activities and good res-
                  Park. He started working for Mike Frank-                taurants, and sees Providence as a small
                  lin, a database professor, who was soon                 city but an eclectic one. Gamze loves it
                  lured to Berkeley. Then in the middle of                and is a big fan of the “Providence” TV se-
                  his third year Ugur changed advisors and                ries. She is an expert on Turkish cuisine,
                  topic area and started studying distrib-                enjoys eating out and has a weakness for
                  uted systems under Pete Keleher. Ugur’s                 dessert! Fortunately, she can burn off the
                  current research explores data manage-                  extra calories by indulging in her other
                  ment issues for advanced distributed sys-               favorite occupations, playing tennis and
                  tems, such as mobile databases, sensor                  working out. They are both big sushi fans
                  networks, and peer-to-peer systems. His                 and are eager to learn about the best
                  broad goal is to design and build scalable              sushi restaurants in town. Gamze is an
                  algorithms and infrastructures that will                amateur Turkish folk dancer; she hopes
                  efficiently support next-generation dis-                 to keep up with her dancing via her con-
                  tributed applications and services. He                  nection with Andy van Dam’s son-in-law,
                  will be teaching a topics course on mobile              who is a professional folk dancer.
                  and ubiquitous computing in the spring
                                                                          Although they still feel they’re in transi-
                                                                          tion mode with all their possessions in
                  Ugur is a huge sports buff and a great                  Maryland, they’re looking forward to set-
                  supporter of the soccer team Fener-                     tling down together in their new Provi-
                  bahce—“Go yellow canaries!” Besides be-                 dence apartment and becoming real
                  ing captain of his college soccer team, he              Rhode Islanders.

                                                                 conduit! 9
                            EFFICIENT LOW-COST
                          DATA AND TRANSACTIONS
                       INTRODUCTION                             For more information, visit http://www.-
                       Security is an essential re-
                       quirement of distributed                 EXAMPLE APPLICATION
                       platforms for business ap-               SCENARIO
                       plications. We address the
                       problem of authenticating                Imagine a trusted financial portal service
                       high volumes of data and                 (such as Yahoo! Finance) that provides
                       transactions     in    non-              dynamic and up-to-date financial infor-
                       trusted distributed envi-                mation, including corporate and news
                       ronments. Current solu-                  feeds, market quotes, and online trading.
                       tions are typically:                     Some key characteristics of this distrib-
                       ❁ Centralized—and there-                 uted application are:
                       fore subject to network de-              ❁ High volume of accesses, queries, and
                       lay and denial-of-service                transactions
                                                                ❁ Widely distributed clients (e.g., geo-
                       ❁ Expensive—to build,
Roberto Tamassia maintain, and operate be-                      graphically dispersed and with varying
                                                                access bandwidth)
        cause trusted data must be maintained in
        a secure environment, guarded 24/7                      ❁ Information originating from various
        ❁ Non-scalable—with limited throughput                  sources outside the portal (e.g., stock ex-
        due to operating and economic con-                      changes, government agencies, and corpo-
        straints                                                rate investor relations offices)

        We are developing a high-throughput sys-                How can such a portal become fully
        tem for authenticating data and transac-                trusted so that the integrity of all data
        tions in non-trusted environments, at the               and transactions is guaranteed authentic
        network edge and outside the firewall. We                end-to-end? Today, the economics are
        use a unique, patent-pending approach                   daunting and the implementation would
        for wide distribution of authentication in-             be largely impractical. For example,
        formation. As a result, we can dramati-                 HTML pages may be dynamically con-
        cally lower the cost of authentication in               structed from several frequently chang-
        such applications as wireless authoriza-                ing information sources, based on user
        tion, B2B e-commerce exchanges, distrib-                requests and user profiles. The volume
        uted storage, end-to-end integrity, tamper              and diversity of such requests make cre-
        detection, and certificate revocation                    ating individual time-stamped digital fin-
        checking.                                               gerprints for any given page very difficult
                                                                and time-consuming. Common techniques
        This work is being conducted in collabora-
                                                                for load balancing, replication and multi-
        tion with AlgoMagic Technologies, Johns
                                                                hosting subdivisions of a web site make
        Hopkins University, and the University of
                                                                using previously available trust ap-
        California, Irvine, and is supported in
                                                                proaches unrealistic, since access to sen-
        part by a grant from the Defense Ad-
                                                                sitive data must be limited and strictly
        vanced Research Projects Agency. My col-
        laborators Steve Baron, Robert Cohen,
        and Rich Sneider at AlgoMagic Technolo-                 Our distributed authentication approach
        gies and Mike Goodrich at the University                offers significant new capabilities and
        of California, Irvine, contributed to writ-             could dramatically change the economics
        ing this article.                                       of authentication for distributed business

                                                      conduit! 10
applications. Highlights of our approach             Previous Approaches
                                                     The simplest approach to implementing
❁ It works with and leverages existing               an authenticated dictionary is to make
Internet infrastructure, protocols and               the responders trusted parties. This
computing platforms                                  trusted-responder approach is used,
❁ It is highly scalable in terms of both             for example, in the online certificate sta-
computational and economic cost                      tus checking protocol (OSCP), where
                                                     trusted OCSP responders answer certifi-
❁ It provides succinct and timely authen-            cate revocation queries posed by clients.
ticity proofs for entire collections of data         The main disadvantage of the trusted-re-
originating from diverse sources                     sponder approach is that each responder
❁ It widely distributes authentication in-           must be placed in a secure location, with
formation to non-secure replication loca-            ensuing large operational costs. Thus,
tions, while strictly maintaining trust              this approach is not scalable for economic
❁ It responds to authentication requests
by clients with minimal network latency              An alternative approach entails having
and computational cost                               the repository periodically sign a finger-
                                                     print of the current version of the entire
We have developed a distributed authen-
                                                     database. The database itself, together
tication system called Secure Transac-
                                                     with the signed, time-stamped finger-
tion Management System (STMS). A
                                                     print, is then sent to the client as a proof
fully operational prototype of STMS has
                                                     of the answer. This database-forward-
been implemented.
                                                     ing approach is used, for example, to
                                                     authenticate certificate revocation lists.
                                                     The database-forwarding approach allows
STMS TECHNOLOGY                                      non-trusted responders and thus has
Authenticated Dictionaries                           lower operational costs. However, it is
                                                     computationally demanding for the client,
The computing abstraction under-
lying STMS is a data structure
called an authenticated dictio-          “when a client makes a
nary, which is a system for distrib-
uting data and supporting authen-
                                       query to an STMS responder,
ticated responses to queries about       it gets back not only an
the data.
                                        answer but also a proof of
In an authenticated dictionary, the
data originates at a secure central             the answer”
site (the repository) and is distrib-
uted to servers scattered around the net-            which needs to process the entire data-
work (responders). The responders                    base in order to validate an answer. Also,
answer queries about the data made by                it is not scalable for communication rea-
clients on behalf of the repository.                 sons, since sending the entire database
                                                     together with the answer uses consider-
It is desirable to delegate query answer-
                                                     able network bandwidth.
ing to the responders for two primary rea-
sons:                                                The STMS Approach
1. It is undesirable that the repository             The main feature of STMS is that it
provide services directly on the network             maintains trust even when responders
due to risks such as denial-of-service at-           are located in insecure, non-trusted loca-
tacks                                                tions. That is, when a client makes a
2. The large volume and diverse geo-                 query to an STMS responder, it gets back
graphic origination of the queries require           not only an answer but also a proof of the
a distributed system of servers to provide           answer. The client can easily validate the
responses efficiently (much like DNS)                 answer and determine that the responder
                                                     has not been tampered with, while rely-
                                                     ing solely on trusted statements signed

                                           conduit! 11
by the repository. The design of STMS al-           ❁ Query formulation by a client
lows untrusted responders to provide ver-
                                                    ❁ Proof verification by a client
ifiable authentication services on behalf
of a trusted repository. This unintuitive           The key benefits to the STMS approach
yet mathematically provable fact is the             include:
key to achieving cost effectiveness.
                                                    ❁ Answers given by the responders are as
The figure below shows a high-level view             trustworthy as if they came from the re-
of the STMS parties and protocol. The re-           pository
pository sends periodic updates to the re-
                                                    ❁ Data is distributed close to the clients,
sponders together with a special signed
                                                    minimizing network delays
                                                    ❁ Deploying responders is inexpensive
                                                    ❁ The repository is protected from risks
                                                    such as denial-of-service attacks
                                                    An additional unique advantage of STMS
                                                    is that it supports historical persist-
                                                    ence; that is, one can perform queries on
                                                    past versions of the database, which
                                                    makes possible a variety of nonrepudia-
                                                    tion applications.
time-stamped fingerprint of the database             STMS comes with Java and C++ toolkits
called the basis. A responder replies to a          for building clients, including modules for
query with an authenticated response,               validating responses. For maximum reli-
consisting of the answer to the query, the          ability and security, cryptographic func-
proof of the answer and the basis. Infor-           tions are performed with the available
mally speaking, the proof is a “partial fin-         standard cryptographic libraries of Java
gerprint” of the database that, combined            and Windows. We also provide an XML-
with the subject of the query, should yield         based protocol for querying a responder
the fingerprint of the entire database. A            over an HTTP connection. We are cur-
proof consists of a very small amount of            rently working on a Web services ap-
data (less than 300 bytes for most appli-           proach in which STMS is accessed
cations) and can be validated quickly. The          through the SOAP protocol.
client finally evaluates the risk associated
with trusting the answer using the fresh-           The table on the opposite page compares
ness of the time-stamp.                             the STMS approach with the trusted-re-
                                                    sponder and database-forwarding ap-
Our patent-pending algorithms, which                proaches with respect to operational cost,
employ only standard cryptographic func-            run-time performance, scalability, and
tions for signing and hashing, guarantee            support for historical persistence.
that the following tasks can be performed
in near real time and with minimal use of           APPLICATIONS
computing resources:                                As a foundation for delivering a wide
❁ Creation of the new signed basis by the           range of trust services, STMS brings au-
repository for each time quantum (the ba-           thentication to the network edge and can
sis is incrementally updated)                       be viewed as a distributed low-cost “au-
                                                    thentication cache” for high-volume
❁ Update of the database copy residing at           transactions. Classes of trust applications
a responder for each time quantum (the              in which STMS can be exploited include:
copy is incrementally updated)
                                                    ❁ User authentication and access control
❁ Assembly of the proof by a responder              for large-scale corporate portals, private
for each query (the proof is obtained by            exchanges, and Virtual Private Net-
combining precomputed partial proofs)               works—User signon can be accomplished

                                          conduit! 12
                      without the need to contact a remote cen-                    ❁ Secure DNS—Host name to IP address
                      tral server.                                                 mappings could be upgraded to include
                      ❁ Wireless two-way authentication—Mo-                        digital signatures. The data-distribution
                      bile device identity can be confirmed by                      approach would be essentially similar to
                      gateways, network services, and applica-                     that used today. Widely distributed
                      tions, and devices can also confirm the                       trusted DNS servers (as STMS respond-
                      identity of selected network services and                    ers) would not have to be located in se-
                      applications to establish two-way trust.                     cure settings.
                      ❁ Trusted XML Web Services—Two-way                           ❁ Tamper detection for Web sites—Web
                      trust can be established between the par-                    page digital fingerprints can be made
                      ties involved in a “chain” of Web Services.                  available as a reference. The computa-
                      Keys for signed or encrypted portions of                     tional load and delay of signing every
                      XML can be obtained from widely distrib-                     page on each web server at all replicated
                      uted servers (STMS responders) rather
                                                                                   locations are avoided.
                      than centralized systems.
                      ❁ Certificate status checking (e.g.,            “Access rights, roles, and
                      for SSL, secure email, and code cer-
                      tificates)—Identity and status can             derived trust assertions can
                      be mutually verified by both sender
                      and receiver of email, publisher and
                                                                    be quickly checked against
                      user of code components, client and           a localized store of consoli-
                      server in a secure dialogue.
                      ❁ Validation services for business
                                                                        dated credentials”
                      partner networks (e.g., healthcare
                      providers)—Access rights, roles, and de-                     ❁ Tamper detection for operating sys-
                      rived trust assertions can be quickly                        tems—System file fingerprints can be
                      checked against a localized store of con-                    made available as a reference on a local-
                      solidated credentials. Public networks                       ized basis for large groups of client sys-
                      can be employed instead of incurring the                     tems.
                      cost of a private value-added network.                       ❁ Tamper detection and license checking
                      ❁ End-to-end integrity for content collec-                   for software applications—Software is in-
                      tion and distribution systems (the exam-                     creasingly distributed or accessed over
                      ple scenario presented at the beginning of                   the network. Metered access can be effec-
                      this article).                                               tively implemented, identity and status of
                                                                                   servers and clients can be mutually con-
                      ❁ Non-repudiation services for electronic                    firmed, and component fingerprints can
                      commerce—Third-party verification and                         be quickly verified.
                      audit capabilities can be implemented to
                      support large-scale B2B networks (e.g.,
                      corporate portals, exchanges, trading sys-
                      tems), as well as high-volume B2C sys-
                      tems (e.g., media distribution, digital
                      tickets, electronic payment systems).

                        Cost   Performance   Scalability   Persistence

 Trusted-responder      High      High          Low            No

Database-forwarding      Low      Low           Low            No

      STMS               Low      High          High          Yes

                                                                         conduit! 13
                                  ALUMNI EMAIL
                           DILIP                       to having someone tell me someday that I
                                                       write “elegantly”.)
                           MS ’84                      All this, in many ways, is where this book
                                                       of mine came from. Too much happens
                          In August, Pen-              around us that is based on unquestioned
                          guin India pub-              stereotypes people hold about each other.
                          lished my first               One such hounds denotified tribes: the
                          book:   Branded              impression that they are all vicious crimi-
                          by    Law    (see            nals. In reality, they are not particularly
                          www.penguin-                 more (or less) criminal than anyone else,
                              but that means little. So via the experi-
books/aspBookDetail.asp?ID=4567). It’s                 ence of these forgotten people, I try to get
about some of India’s most forgotten peo-              my readers to ask questions about preju-
ple—its denotified tribes, once actually                dice, to ask questions above all of them-
defined as criminal. But more than that, it             selves. I try to do this with those tools I
is also a look at prejudice, a phenomenon              first learned about in CS: reason, clarity,
that shapes their lives every day and also             cut the fluff.
greatly interests me for various reasons.
                                                       And the interesting thing is that while a
Now that it’s out, I get asked all the time:           lot of people see it as startling, this appar-
CS, but now you write? What’s the story?               ent “switch” from CS to writing, I see it as
      I came to writing late, true. It never           a smooth and even logical transition. It’s
      occurred to me as a possible career              almost as if I write, and write the way I
      in my university years, nor in my                do, because of my years in CS. I owe that
      years in software. But when I re-                to Brown and to several people who fea-
      turned to India in 1992, I tried writ-           ture regularly in these pages. So please
      ing for a lark. Got published a few              read my book! Tell me what you think.
      times. Wrote some more, then more      
      regularly. And pretty soon it was
      clear to me that I had found what I
      truly wanted to do with my life.
                                                       RICHARD HUGHEY, PhD ’91
      Write.                                           Hi all! This past year I have bumped into
      Yet I never regretted my time in CS,             a few Brown folks. The most surprising
      because it strikes me as marvelous               was at a July computational biology con-
      training for writing. Not everybody              ference (ISMB) overhearing talk of rub-
      would take this route, I know. But               ber chickens. On turning around, I
      the lessons from CS serve me well                mentioned having one in my office (one
      when I write. At its best, and I’m               decaying original rubber chicken, and
      sure conduit!’s readers will agree,              three plastic replacement juggling chick-
      programming is an intensely cre-                 ens!). The speaker turned out to be Sonia
ative process. Good software is every bit              Leach, not only a Brown student but also
as satisfying and elegant as a good piece              occupying my old office—555. We chatted
of writing. And of course the word “ele-               about what we were doing, and I was able
gant” is one all programmers aspire to.                to clear up her puzzlement about how a
Software described that way is an exhila-              VLSI chip plot wound up in the window of
rating mix of functionality, leanness, clear           555—it’s from my dissertation. I’ve gone
thought and a certain beauty.                          on to supervise the design, construction,
                                                       and use of another parallel processor at
I try to write like that: clearly, simply. Cut         UC Santa Cruz.
to the core of issues, find connections, be
sure of myself. These are the attributes of            Ethan Miller (ScB ’87) joined our CS de-
writing that I value, that I aspire to. (Also          partment this year; I remember seeing

                                             conduit! 14
his name a few times as a grad student,                  ing Silicon Valley tech economy, I began
perhaps in 169 (operating systems).                      working for Tufts University Hillel in
                                                         Boston, and then a year later moved to
Being an involved graduate student has
                                                         UC Berkeley Hillel in California, where I
caught up with me, and now I’m chair of
                                                         was earning a non-profit salary living just
our computer engineering department. I
                                                         above the poverty line. Hillel is the inter-
ran into Tom Dean and Jeff Vitter at the
                                                         national organization that serves the
CRA meeting last March as a pre-chair
                                                         needs of Jewish college students (I logged
and, not ducking quickly enough, the po-
                                                         many hours as a student at the Brown
sition caught up with me two months ago.
                                                         Hillel when I wasn’t in the CIT), and in
I won’t go into all the pleasures of the po-
                                                         my misleadingly titled job of “Program
sition, as then no graduate student would
                                                         Director” I spent my days meeting Jewish
ever get involved in anything for fear of
                                                         students, planning and attending fun
eventually becoming a department chair!!
                                                         events with them, and playing my guitar.
Parts of the job, like the planning of re-               In addition, because of my extensive com-
search and academic programs, are quite                  puter science background, I was also reg-
fun—CE has five positions this year,                      ularly called upon to teach my co-workers
starting the ramp-up to UC’s “Tidal Wave                 how to perform highly technical com-
II” enrollment surge. I’ve also been devel-              puter-related tasks such as: sending
oping BS (starting 2001-2), MS and PhD                   group emails, copying files to and from
(starting 2002-3, we hope) curricula for                 our local area network, replacing printer
Bioinformatics degrees with my col-                      toner cartridges, and of course, program-
leagues Kevin Karplus (protein structure                 ming the VCR. All in all, the job was a
prediction) and David Haussler (putting                  great experience for me.
together the human genome, among
other things). We’re planning to build                   I enjoy working with people, and find that
this into a department in a couple of                    I fit in comfortably in the non-profit
years, and are currently recruiting three                scene, so with that in mind, I have de-
faculty in bioinformatics and related ar-                cided to go back to school to brush up on
eas.                                                     my management skills before trying my
                                                         luck as a director of my own non-profit or-
Otherwise, Santa Cruz is a great place—
                                                         ganization. So, I have just recently left
with a campus of redwoods, I quickly
                                                         Berkeley and moved to Evanston, Illinois,
overcame my trepidation at interviewing
                                                         where I will be starting as a business stu-
on the west coast!
                                                         dent at the Kellogg/Northwestern School
                                                         of Management. Ironically, it seems that
JOSH MILLER, AB ’96                                      a large number of my classmates have CS
                                                         undergrad degrees and are former dot-
Hello out there to the Brown CS Commu-                   commers who decided to sell their BMWs
nity—Well, I decided to mail a little                    and apply to business school when the
change-of-address notice to Suzi Howe                    economy collapsed and their companies
and she craftily cornered me and asked                   went under! I guess the old adage is true,
me to submit a quick update to conduit!                  “If you don’t know where you’re going,
on what I’ve been up to, not realizing that              any path will take you there.”
such a request would elicit a dual confes-
sion of unprecedented proportions. So                    Despite my unusual path, I have always
here it goes: 1) I haven’t done anything                 been glad I went the CS route at Brown,
related to computer science since I gradu-               and in a true liberal arts spirit I believe
ated and 2) despite that, I still read my                that my whole CS experience—TAing,
conduit! cover to cover (including the                   group work, and even building computer
technical articles) every time it comes in               systems—was a great educational prepa-
the mail. Weird, eh? So what exactly have                ration for my work since graduating. I’d
I been up to? Well, in 1996, when most of                love to hear from folks, and encourage
my graduating class (including the phi-                  other Brown CSers on nontraditional
losophy majors) went straight from                       routes to write in and tell their stories!
Brown to cushy jobs working in the boom-                 Josh Miller,

                                               conduit! 15
                                                                                past work and education experience. Lo
                  JON MOTER, ScB ’99                                            and behold, it turns out he graduated
                  Spike forwarded the following, which Jon                      from Brown in ’94 with an Art/Semiotics
                  Moter posted to brown.cs.stupid:                              degree. Apparently he “graduated cum
                                                                                laude, with a performing arts thesis that
                  With Microsoft’s new version of Office,                        involved twisting myself into a represen-
                  they’re finally killing off the annoying                       tation of Michelangelo’s David.” Always
                  Paperclip Assistant. As a marketing                           good to see Brown alums in prominent
                  move to push MS Office XP, they’ve cre-                        positions. Jon “You still have an ac-
                  ated a site for Clippy, bemoaning his own                     count???” Moter.
                  fate and looking for a job. It’s actually
                  fairly    amusing:                    Said Spike: “So...even though the paper-
         He has a resume (http://                          clip’s not from Brown CS, perhaps it de-
         with                         serves mention in conduit! (or to our

                                                                                SCOTT SMOLKA, PhD ’84
                                                                                Thought this photo might be of interest
                                                                                to conduit!’s readers: three of Paris’s
                                                                                Ph.D. students: me, Alex Shvartzman
                                                                                Ph.D. ’92 and Dina Goldin Ph.D. ’97. I
                                                                                was in Boston visiting Dina; we’re work-
                                                                                ing on a joint paper with Peter Wegner
                                                                                about Turing machines, transition sys-
                                                                                tems and interaction. Alex was visiting
                                                                                Nancy Lynch’s group at MIT. The restau-
                                                                                rant is somewhere in Kendall Square. All
                                                                                the best, Scott.
     l to r, Scott Smolka, Alex Shvartzman and Dina Goldin

                                                                            ceive the world outside them. The 27th
                                                                            Industrial Partners Program Symposium,
       27th IPP SYMPOSIUM                                                   held last May, brought together leading
    VISION-BASED INTERFACES                                                 researchers from companies that are at-
                                                                            tempting to revolutionize the interaction
                                                                            between machines and humans by en-
                           The paradigm of human-com-
                                                                            dowing computers with perceptual capa-
                           puter interaction based on win-
                                                                            bilities. Speakers from Microsoft, Intel,
                           dows, icons, menus, and pointing
                                                                            Compaq, Mitsubishi, and IBM spoke
                           devices has changed little since
                                                                            about the fundamental science needed to
                           its development in the early
                           1970s. With the exception of                     enable computers to “see” and under-
                           some limited voice-recognition                   stand their users. They also presented
                           technology, a computer’s knowl-                  some novel “perceptual user interfaces”
                           edge of the external world comes                 that explore how these abilities might
                           through this screen/keyboard in-                 change the way computers and humans
                           terface. In contrast, one-on-one                 interact. We first heard about the “Ea-
                           human interactions involve a                     syLiving” project from Dr. John Krumm
                           great deal of context in which                   of Microsoft Research. At their Red-
                           the speakers are aware of their                  mond research facility, Microsoft has built
                           shared environment and the ob-                   a “living room” with computer screens
                           jects in it as well as each other’s              embedded in the walls that can provide
Symposium host,   facial expressions and gestures.                          information and entertainment (figure on
 Michael Black                                                              next page). The room is also equipped
                  This sort of rich interaction is not possible
                  with today’s computers due to their lim-                  with cameras and other sensors that de-
                  ited intelligence and their inability to per-             tect the location of people in the room, de-

                                                                  conduit! 16
      IPP Symposum speakers from l to r: John Krumm, Microsoft Research; Michael
      Black, Brown; Matthew Brand, MERL; John MacCormick, Compaq; Jim Rehg,
         Compaq; Myron Flickner, IBM Almaden; Gary Bradski, Intel Corporation
termine their identity, and track them as              room at one time. These features provide
they move about. The project is exploring              a testbed for exploring novel interfaces
the kinds of services that might be sup-               for product groups at Microsoft.
                                                       Reliable detection and tracking of people
                                                       in complex, changing environments such
                                                       as the living room of the EasyLiving pro-
                                                       ject is still an open problem. Dr. John
                                                       MacCormick of Compaq’s Systems
                                                       Research Center spoke about their
                                                       probabilistic methods for people tracking.
                                                       Their BraMBLe system builds probabilis-
                                                       tic models for what the background
                                                       (room) and foreground (people) look like.
                                                       These models are used in an elegant
                                                       probabilistic formulation that exploits
                                                       Bayesian probability theory to combine
                                                       information measured from a camera
     Microsoft Research, EasyLiving
                                                       with past measurements and prior
  Project. This “intelligent” living room
                                                       knowledge of the world. The result is a
           senses people in it
ported by this sort of “intelligent room.”
For example, the room can “learn” indi-
vidual preferences that are used to
change the lighting, heating, or music se-
lections. Additionally, when a known per-
son enters the room, their computing
environment follows them with their
email, bookmarks, and instant messages.
Supporting these applications are a num-
ber of cameras that compute the three-di-
mensional structure of the room and
detect the location of people. A tracking
system combines information from the                       Compaq System Research Center.
cameras and pressure sensors in the fur-                    Probabilistic tracking of multiple
niture to track multiple people in the                           objects in a scene

                                             conduit! 17
                         “real-time” tracker (figure above) that can               rived. He showed impressive results of
                         detect multiple people from a single cam-                tracking a face in a monocular video se-
                         era, determine how many people are in                    quence while recovering its three-dimen-
                         the scene, and track them even when they                 sional shape and motion (figure below).
                         partially occlude each other.
                         Dr. Jim Rehg talked about related work
                         at Compaq’s Cambridge Research
                         Laboratory that uses body tracking in a
                         “smart kiosk.” This kiosk uses video cam-
                         eras to detect people that are near it and
                         track their motion (figure at left). A dis-
                         play screen has a realistic-looking human
                         face that talks to people and encourages
                                                                                          Mitsubishi Electric Research
                         them to approach the kiosk. An installa-
                                                                                        Labs. 3D facial motion tracking
                         tion of the kiosk in the Cybersmith Cafe
                                                                                               with uncertainty
                         in Harvard Square allowed Compaq to as-
                         sess the potential market for kiosks with                The performance appears more robust
                         sensing capabilities. This installation also             than previous systems and the informa-
                         let Compaq experiment with different fea-                tion derived from the motion estimation
                         tures and evaluate this type of perceptual               can be used for both expression recogni-
                         user interface. The researchers observed                 tion and the animation of 3D graphics
                         that the interactive talking kiosk drew                  models of faces.
                         people to it and that high-quality content               The motion of the eyes is of particular in-
                         was the most important factor in holding                 terest in human-computer interfaces. Dr.
                         their attention. Furthermore, the experi-                Myron Flickner of the IBM Almaden
                         ment revealed that interactive entertain-                Research Center described their “Blue
                         ment was more compelling in this                         Eyes” project for detecting and tracking
                         application than information services. To                eyes using infrared light and a video cam-
Compaq Cambridge improve the interaction with the kiosk,                          era (figure below, left). By locating the pu-
Research Laboratory. Dr. Rehg’s recent research is focusing on                    pil and reflections on the cornea, they are
Smart kiosk installed in probabilistic methods for detecting                      able to track the human gaze accurately.
  Cybersmith Cafe        whether or not a person in front of a cam-
                                                                                  To explore new perceptual interfaces,
                         era is actually speaking by analyzing the
                                                                                  IBM’s new robot Pong can detect faces
                         motion in the video in conjunction with
                                                                                  and facial expressions as well as express
                         the auditory signal.
                                                                                  them (figure below, right). Pong has
                         While the above systems focused prima-                   speech recognition and synthesis software
                         rily on detecting and tracking people as                 that lets it interact with a user while a
                         crude “blobs,” many types of human-to-                   camera hidden in its nose detects faces
                         human interaction exploit detailed infor-                and analyzes facial expressions. Dr. Flick-
                         mation about the face and hands.
                         Recognizing information about fa-
                         cial expression requires an analysis
                         of facial motion, which is challeng-
                         ing due to the complex deformations
                         the human face can undergo. Dr.
                         Matthew Brand of the Mitsub-
                         ishi Electric Research Labora-
                         tory (MERL) described a novel
                         mathematical formulation of this
                         motion estimation problem that
                         uses matrix algebra to propagate
                         uncertainties in image measure-
                         ments in such a way that informa-
                         tion loss is minimized and optimal                IBM Almaden Research. l, BlueEyes gaze
                         estimates of the facial motion can be de-               tracking; r. lifelike Pong robot

                                                                        conduit! 18
                      ner gave a live demonstration of Pong                 While the form of future interfaces can-
                      and talked about other projects at IBM                not be predicted, it is clear that many of
                      exploring “attentive” user interfaces that            our partner companies believe that they
                      combine physiological measurements,                   will change radically and that they will
                      eye gaze, and body language that they                 exploit new perceptual capabilities. In
                      hope will make computers more widely                  particular, this workshop explored com-
                      useful.                                               puter vision technologies that might give
                                                                            computers information about the visual
                                 The above methods have one
                                                                            world occupied by their users. With the
                                 simple thing in common that
                                                                            active research on this problem in aca-
                                 distinguishes them from tra-
                                                                            demia and industry, it is a fertile area for
                                 ditional interfaces—the visual
                                                                            partnerships like those supported by IPP.
                                 processing is computationally
                                 intensive. Intel is in the busi-           For more information about these pro-
                                 ness of selling computation, so            jects, see:
                                 their interest in these new                Microsoft, EasyLiving project: http://re-
                                 perceptual interfaces is not     
                                 surprising. Dr. Gary Brad-                 Compaq, Smart Kiosk:
                                 ski from Intel’s Micropro-       
                                 cessor      Research      Labs             fault.htm
                                 presented their efforts to de-             MERL, facial motion: http://www.merl.-
                                 velop an open source com-                  com/people/brand/
                                 puter vision library to support
                                                                            IBM, Blue Eyes http://www.almaden.-
                                 the growth of novel applica-     
                                 tions. This library of vision
                                 software provides a huge                   IBM, Pong
                                 number of basic and advanced
                                 methods that support every-                Intel, Open Source Computer Vision Li-
                                 thing from low-level segmen-               brary:
Intel. Human activity analysis tation to full-body tracking and             mrl/research/opencv/
 from the Intel Open Source probabilistic action recognition
   Computer Vision Library     (figure at left).

                                                                            In response, Matthias Felleisen, Corky
      HOW TO CREATE THE BEST                                                Cartwright and their students (including
                                                                            me) built a new curriculum for introduc-
        OO PROGRAMMERS                                                      tory programming. Like Abelson and
                                                                            Sussman, we steered clear of the nagging
                      Got you to look, didn’t I? I promise I’ll get         terminology and low-level details that
                      to this subject at the end. Maybe you                 dog so many introductory programming
                      want to just jump right there—go ahead!               texts. Unlike them, we stressed the sys-
                                                                            tematic construction of programs from
                      BACKGROUND                                            crisp and accurate definitions of data—
                      The TeachScheme! Project was conceived                the stuff of OOP, in other words, without
                      in frustration. Our group at Rice Univer-             the verbiage and overhead of OO. This let
                      sity had spent several years overhauling              us take students with no prior program-
                      Rice’s introductory curriculum. Rice had              ming experience through all the interest-
                      adopted an approach based on Abelson                  ing topics—lists, trees and other useful
                      and Sussman’s seminal work at MIT but,                data types, higher-order functions and
                      despite its sublime content, it wasn’t do-            non-local control, as well as the routine
                      ing the job for us: it provides immense in-           stuff of scope and mutation—with a very
                      sight into structuring systems but, we                particular emphasis on program design.
                      felt, gave little insight into structuring
                      programs.                                             Over the years, three features have
                                                                            driven our success.

                                                                  conduit! 19
           When asked to elaborate on some jpegs of him windsurfing, Tom Dean
           said, “You can ask Keith Hall and Stu Andrews. I walked down to the
           beach a few blocks from our house on a Saturday a couple of weeks back
           and found Keith and Stu throwing a frisbee, with Stu’s windsurfer lying
                                                  nearby. Stu asked me if I wanted
                                                  to give it a try and I said sure. I
                                                  used to do quite a bit of wind-
                                                  surfing, having planned vaca-
                                                  tions around wind-surfing at
                                                  Maui, Columbia River Gorge,
                                                  the Sea of Cortez in Baja Mexico
                                                  and numerous other windsurf-
                                                  ing meccas. My short stint on
                                                  Stu’s board was the second time
                                                  I’ve windsurfed in three years. I
                                                  was lousy but Stu and Keith
                                                  were supportive.”
                                                         Keith is a fourth-year PhD stu-
                                                         dent, Stu is in his third year.

1. We developed a methodical response to             2. We built a beginner-friendly program-
that most frustrating of questions, “Can             ming environment, DrScheme (“Doctor
you help me find my bug?” Our work pre-               Scheme”). DrScheme is an outgrowth of
sents a six-step design recipe that begins           several years of observing how students
with an understanding of data, then pro-             program in labs. Its innovations are the
                                                     fruit of several PhD theses. From a peda-
ceeds to understanding the black-box be-
                                                     gogic perspective, one feature towers over
havior of the function, then to deriving its
                                                     others. It exposes the fundamental mis-
template from the data. Only in the fifth
                                                     match between texts, which provide a
of the six steps does the student actually           stratified view of languages, and environ-
write the function. (The last, in case               ments, which toss the student into a lin-
you’re wondering, is testing.)                       guistic mass for which they’re often
                                                     completely unprepared. DrScheme in-
This methodology not only gave our TAs
                                                     stead provides language levels that a stu-
and us a grading rubric, it also gave stu-           dent can increment as her skills grow. In
dents a valuable debugging checklist. To             particular, each language level is careful
our surprise, we found that most student             to provide feedback using only the lan-
errors had little to do with their code:             guage and terminology a student is ex-
their errors were at a much earlier phase,           pected to know at that level. In contrast,
usually in defining the structure of their            pretend you’re in your first week as a pro-
data. If you don’t understand the data you           gramming student, type ‘wage * hours
have, you obviously can’t hope to process            = salary’ into a C++ environment, and
it very well. We suddenly developed a                make sense of its response. (We’ve ob-
much more sophisticated understanding                served high-school students nearly re-
of student errors (a big help when plan-             duced to tears from trying to decipher
ning what to revisit in lecture, for in-             repeated feedback of this sort.)
stance). For their part, students now had            3. We exploited the multiplatform, graphi-
a much more useful debugging methodol-               cal nature of DrScheme to build numer-
ogy than making random syntactic                     ous libraries to support interesting
changes and putzing about in a debugger.             exercises. The libraries help students

                                           conduit! 20
write simple games, build very rudimen-              school teachers, we realized why these
tary animations (no, no, not like Andy’s or          students had these views. The AP CS ma-
Spike’s!), try a little Web programming,             terial has a lowest-common-denominator
create file-system managers, and so on.               feel; indeed, almost uniquely among the
The libraries plug into our Scheme’s                 AP disciplines, many leading universities
graphics and systems APIs, but students              (including Brown and Rice) don’t recog-
write all the critical control elements.             nize scores in this subject. This lack of
Most of these are extended exercises that            recognition stems, in part, from a huge
grow in complexity, letting students peel            disparity in depth between the high
away layer upon layer. In short, we ex-              school and college levels.
pose them to two important software en-              Despite the best intentions of its framers
gineering      principles—model-view-con-            to remain at a principled level, we found
troller designs and iterative refinement—             that teachers in the field were forced to
all while they’re having fun!                        deal with a myriad of low-level details of
                                                     syntax and machine organization, to the
THE EARLY DAYS                                       point where most never got to useful data
We’ve taught this material since 1994 at             structures or other concepts. Many teach-
Rice, and it’s been a great success. Enroll-         ers reported that material we’d consid-
ments skyrock-                                                                ered     routine—
eted;     perhaps                                                             programming
more        impor-
                     “Enrollments skyrocketed;                                over trees, for in-
tantly, the num-     perhaps more importantly,                                stance—was pre-
                                                                              sented gingerly,
ber of women
grew dispropor-     the number of women grew                                  at the end of two
tionately. Many                                                               semesters, and
who’d taken pro-
                         disproportionately”                                  only the best stu-
gramming                                                                      dents were ex-
classes in high school (especially women)            pected to get it absolutely right; most
said they enjoyed the more structured                never extracted themselves from the long
and goal-driven curriculum, in contrast to           dark night that is debugging. In contrast,
the seemingly random activity they’d ex-             we considered it a failure if virtually ev-
perienced before. When students mas-                 ery student, aided by the design recipes
tered even our rapid pace of material, we            where necessary, didn’t nail this material.
fried their brains in bonus lectures.                Matthias Felleisen then asked the ques-
One small group of high school students,             tion we hadn’t dare pose: Why can’t we
however, always resisted this material.              teach this material in high schools?
Over time, we came to realize most of                What’s the real difference between a high-
these were students who’d had Advanced               school student and a first-year college stu-
Placement computer science (AP CS)                   dent anyway, save for a summer of frolic?
courses. Most AP students recognized                 (Not our students, who’re too high-minded
that, by about the sixth week, our course            to spend their summers that way, but
was in terrain they found completely un-             maybe someone else’s.) Despite what our
familiar. A small group, however, resisted           viewbooks would like to suggest, students
the call to treat the material at its own            aren’t magically transformed when they
level of abstraction. Common questions               walk through a Van Wickle Gate or a Sal-
and complaints included, “Isn’t that just a          lyport. So why not take this to high
linked list?” “Where’s the null pointer?”            schools?
“Why won’t the number overflow?” “Isn’t
recursion bad for the stack?” “This doesn’t          GROWTH
behave like a computer.” What amazed us              Yep: we had no idea what we were in for.
was not simply how frequently these stu-             We were used to college educators, to
dents complained, but how misinformed                whom you can mail a book with a friendly
they were.                                           note. But high school teachers—especially
When we eventually set to studying the               in computer science—are so overworked,
AP CS material and talking with high                 it’s utterly unfair (and impractical) to

                                           conduit! 21
simply mail them a manuscript. (Com-                   ing how students learned C++ and
puter science professors joke about being              Scheme. Somewhere along the way, we
asked to fix printers; for teachers, it’s part          made a semantic shift: our on-line “lec-
of their job.) In the furious market of the            ture notes” grew into a “book.” Having
late 1990s, many computer science teach-               trained the teachers, we became victims
ers left for industry; the ones remaining              of our limited success: as teachers began
had no training in the discipline at all               to use the material in classes, we were
(and they were teaching C++!). They were               obliged to support them. We had to write
valiantly keeping their classes alive, but             and publish solutions. We began to write
learning a whole new approach, unaided,                a programming environment user’s guide.
was out of the question.
                                                       We added new libraries and exercises. We
In the summer of 1996, therefore, we                   revised the language levels. The codebase
brought in a high-school teacher for a few             grew to over half a million lines.... In
days of intensive training. We had no idea             2000, our demand grew too large to fit
what she knew or how to teach her, and I               into one workshop, so we ran two. We
imagine those few days were pretty rough               drew teachers from as far afield as Mex-
on her. But she survived it, so the next               ico, Saudi Arabia, Switzerland and Japan.
year we brought in two. (Exponential                   (Okay, so maybe Mexico isn’t all that far
growth! We had visions of Malthus.) As                 from Houston.) We also began to benefit
our first big test, in 1998 I recruited a               from the diaspora of PhDs: Matthew Flatt
group of 18 teachers from Texas comput-                ran a small workshop at his new home,
ing conferences to attend a week-long                  the University of Utah.
summer workshop at Rice. Looking back,
we still knew very little about our audi-              This year, the workshops themselves went
ences; it’s a wonder that many of those                national. We ran workshops in four differ-
teachers are still with us. (What doesn’t              ent states, training nearly a hundred
kill us makes us stronger.)                            teachers. The largest, hosting nearly forty
                                                       teachers, took place in July, 2001 at
Suddenly, TeachScheme! had become real.                Brown, run jointly by my colleague, Kathi
We were conducting school visits, observ-              Fisler, of Worcester Polytechnic, and me.

           Grad student Jasminka Hasic played host to a family of house finches during the
           spring/early summer. The pair began building their nest atop the open window of her
           5th floor office; consequently, the window had to remain open come wind and weather.
           In order to discourage the birds from flying into the room and becoming disoriented or
           hurt, Jasminka, herself a newly hatched American citizen, used her giant-sized
           American flag to cover the window and darken the room. The nest proved a remark-
           able, though messy, feat of engineering, teetering on the edge of oblivion. The female
           finch laid four eggs, all of which hatched. The highly vocal babies were pretty dis-
           tracting at feeding time, but very charming. Despite concern that the babies’ first
           attempt at flight would be their last, each one fledged successfully.

                                             conduit! 22
         We, of course, merely soaked up the                 One reason for TeachScheme!’s growing
         credit; our helpers made it work smoothly.          university adoption is its superb conduit
         We had three great returning teachers               to true object-oriented programming.
         and five superb students: three PhD stu-             Corky Cartwright at Rice worked with us
         dents from Brown (Greg Cooper, Rob                  to revamp the second course to teach stu-
         Hunter and Dave Tucker), and one each               dents algorithms and design pattern-
         from Cornell and Northeastern. This stu-            based programming in Java. The beauty
         dent group makes us feel like we’ve taken           of this scheme is that once students have
         the first step towards closing an educa-             internalized the TeachScheme! design
         tional cycle.                                       recipe, it takes just about a day to start
         One of the major attractions of our curric-         producing authentic Java! Scheme easily
         ulum is that it costs teachers, students            morphs into code obeying the Interpreter
         and schools nothing. We distribute soft-            and Composite patterns; their Scheme ex-
         ware and support material free on the               perience with higher-order functions has
         Web. Our text, How to Design Programs,              prepared them well for inheritance-based
         is published in hardback by the MIT                 abstraction; even the Visitor pattern is an
         Press, but we asked to be able to distrib-          easy step away. Students get through
         ute it free on the Web, and they gener-             most of the Gang of Four book, in addition
         ously agreed. Some poorer schools, both in          to the usual algorithmic material, in their
         the U.S. and abroad, use the book entirely          second semester.
         through this medium. Finally, with gener-           A few high-school and college faculty
         ous support from the National Science               who’ve adopted this route are finding that
         Foundation, U.S. Department of Educa-               their students internalize OOP better
         tion, Exxon and CORD, an educational                through this two-step approach than they
         nonprofit that’s adopted our curriculum              did starting with Java. By the time they
         for its Academies of Information Technol-           confront their first ‘public        static
         ogy, we pay expenses and stipends to vir-           void main’ the students are already
         tually all teachers who attend workshops.           masters of useful data structures and sev-
                                                             eral introductory algorithms. Having al-
         OOP                                                 ready learned one language, the second
         We’re (fortunately) lowering the exponent           one comes a lot quicker—because the
         in our growth curve. More importantly,              Scheme we teach maps easily to Java. The
         we’re seeing signs of the change we set             design recipes entirely take the mystery
         out to accomplish. A growing population             out of recursion (sorry!) in a way that no
         of teachers recognizes that there’s an al-          prior approach we know of has. In short,
         ternative to what they teach. Our “repeat           there’s no excuse for not teaching chal-
                                                             lenging and interesting topics.
                                                             People sometimes ask me whether func-
“One reason for TeachScheme!’s                               tional programming isn’t unintuitive; per-
growing university adoption is its                           haps students should really only see it as
                                                             (college) juniors. My evidence to the con-
 superb conduit to true object-                              trary is simple. I point to hundreds of stu-
                                                             dents, starting from as young as the 7th
    oriented programming”                                    grade, at schools big and small, poor and
                                                             rich, in the U.S. and in several foreign
         customer” ratios (teachers who use the
                                                             countries. These students write rich, in-
         curriculum more than once) are huge. Al-
                                                             teresting and even fanciful programs. We
         most every week, we hear from either a
                                                             hope it isn’t just the language: the envi-
         teacher who struggled before who says
                                                             ronment certainly helps, and teachers tell
         he’s thriving now, or one who was comfort-
                                                             us the design recipe wins big, too. But the
         able before but whose students are now
                                                             bottom line is, while functional program-
         building projects beyond her expectations.
                                                             ming may not be intuitive to professors, it
         Perhaps one way to see that Teach-
                                                             sure is to kids. (And it seems to be in-
         Scheme! has come of age is that promi-
                                                             creasingly intuitive to programmers in ev-
         nent universities other than our own,
                                                             erything from JavaScript to Python—as
         both in the U.S. and abroad, have adopted
                                                             their programmers hit on the same prob-
         our curriculum.
                                                             lems functional programmers did two de-

                                                   conduit! 23
cades ago, their languages have rapidly               syntactic details of some language; we
expanded in their functional offerings, as            must train them to harness this beast
numerous articles on these languages at-              called computation. If anything, the non-
test.)                                                majors need it more than the majors. In
                                                      one semester (if that’s all they’ll give us),
THE FUTURE                                            we must teach them principles they will
                                                      recognize for decades to come. To do this,
What’s next? We’re building many more
                                                      our courses have to get at the heart of
extended exercises and improving the
                                                      computation as quickly as possible and
quality and features of DrScheme. These
                                                      stay there as long as they can. People
improvements induce difficult research
                                                      must internalize these ideas so well that
problems behind the scenes, with the ben-
                                                      they recognize them in whatever they
efit (or danger, if you don’t like the atten-
                                                      choose as a concentration.
tion!) that the solution will immediately
download to thousands of desktops in the              Brown is leading the way here. Last year,
next release. The challenge we face as re-            Lisa Cozzens ’01 completed a senior hon-
searchers and programmers is the same                 ors thesis under my supervision in which
one that actors face: always to stay in               she built several extended exercises im-
character. How do you build a powerful                plementing basic processes from biology.
type-inference engine that doesn’t baffle a            That material caught the eye of a new au-
10th grader? Stay tuned!                              dience: biology teachers. Two of them at-
But I think much more is at stake. I’ve be-           tended the workshop at Brown this past
lieved for some years now that computer               summer. Now they’ve begun to teach pro-
science has come to reside at the heart of            gramming to their students, too. The rev-
a true liberal arts curriculum. What mat-             olution has begun.
ters isn’t the dust and noise: the details of        Acknowledgments: Thanks to my part-
word sizes and endianness. Rather, it’s              ners in crime, especially Matthias Fel-
the notion of computation, what Abelson              leisen, Robert Bruce Findler, Matthew
and Sussman called a procedural episte-              Flatt and Kathi Fisler. Thanks also to the
mology. This is why computer science has             other TeachScheme! staff, and to the
gone from bridesmaid (think of those                 teachers who’ve endured us!
early number-crunchers) to guest of honor            URLs: The Project:
(think biology).                           
Our curricula must reflect this shift. We             The Text:
need to give students more than just the   

Michael Black. In the spring Michael                  In July Michael attended the Interna-
organized an Industrial Partners Pro-                 tional Conference on Computer Vision
gram Symposium on Vision-based Inter-                 held in Vancouver. He was one of the Area
faces (details in this issue).                        Chairs for the conference and his stu-
                                                      dents presented two papers. Hedvig
He has recently been awarded two grants:              Sidenbladh presented a paper on “Learn-
an NSF ITR grant on “The Computer Sci-                ing Image Statistics for Bayesian Track-
ence of Biologically Embedded Systems,”               ing,” while Fernando De la Torre talked
(a collaboration with Elie Bienenstock in             about “Robust Principal Component Anal-
Applied Math and John Donoghue in                     ysis for Computer Vision.” Along with
Neuroscience) and an ONR grant to study               David Fleet from Xerox PARC, Michael
“Motion Capture for Statistical Learning              had an invited paper, “Probabilistic De-
of Human Appearance and Motion” as                    tection and Tracking of Motion Bound-
part of the DARPA Human-ID project.                   aries,” in the Distinguished Paper Track

                                            conduit! 24
at the International Joint Conference on                the department and CIS from his hotel
Artificial Intelligence held in Quebec dur-              room. (Tom was appointed interim Vice
ing August.                                             President for Computing and Information
                                                        Services as of July 1; the appointment
During a busy spring of travel, Michael
                                                        runs through next June, at which point he
gave an invited talk on human motion
                                                        hopes to turn over the job to a permanent
tracking at Workshop on the Convergence
of Vision, Video, and Graphics in Berkeley
and talks on brain-computer interfaces at
the Microsoft Research Vision Symposium                                   www
and at the Workshop on Vision-Based Per-
ceptual Interfaces hosted by the Interac-
tive Institute in Stockholm, Sweden. He                 Amy Greenwald. Like last year, one
was back in Sweden in August to give an                 of the highlights of Amy’s summer was
invited talk at the 2001 Stockholm Work-                her participation in the Distributed Men-
shop on Computational Vision, held on an                tor Program, sponsored by the Computing
island in the Stockholm archipelago. Ad-                Research Association. Her mentorees,
ditionally, he gave colloquia on human                  Victoria Manfredi and Julia Farago, vis-
motion analysis at the University of Roch-              ited from Smith College and Harvard
ester, the University of Western Ontario,               University, respectively. In addition, Amy
the University of Pennsylvania, and New                 traveled from eastern to western Canada.
York University.                                        In May, she presented a paper at Autono-
                                                        mous Agents in Montreal. In August, she
This fall Michael is teaching a new course
                                                        chaired a workshop entitled “Economic
on brain-computer interfaces that is
                                                        Agents, Models, and Mechanisms” in Se-
drawing curious students from depart-
                                                        attle, from where she took a backpacking
ments such as physics, neuroscience, ap-
                                                        holiday in British Columbia (a.k.a. BC
plied math, cognitive and linguistic
                                                        a.k.a. Bug City). And in the middle of the
sciences, and engineering, as well as com-
                                                        summer, Amy attended ICML (Interna-
puter science.
                                                        tional Conference on Machine Learning)
                                                        at Williams College with David Gondek
                  www                                   and Keith Hall, both PhD students in
                                                        their fourth year.
Eugene Charniak. Eugene presented
his paper “Immediate Head Parsers for
Language Modeling” at the 2001 Associa-
tion for Computational Linguistics (ACL)
Conference, this year held in Toulouse.                 Shriram Krishnamurthi. In addition to
Eugene’s technical article in the last con-             having the usual amount of fun, Shriram
duit! was based on this paper. The ACL                  ran a successful TeachScheme! workshop
conference is generally considered to be                at Brown (see article on page 19). He and
the top conference in this area, and this               Kathi Fisler of WPI hosted nearly 40 high
year for the first time it awarded a “best               school teachers from around the country
paper” award. Eugene’s paper shared the                 and abroad, training them in an innova-
award with one other group.                             tive computer science curriculum. He has
                                                        more at
Tom Dean. Tom spent half of August in
Seattle, attending meetings of the board                David Laidlaw. David has been award-
of trustees for IJCAI Inc., giving talks at             ed an NSF CAREER grant, the agency’s
IJCAI-related and satellite meetings and                most prestigious award for junior faculty.
workshops including an invited talk at                  His work under this grant will focus on
ATAL-2001 (Agent Theories, Architec-                    shape modeling and its applications cou-
tures and Languages), and trying to run                 pled with the development of a methodol-

                                              conduit! 25
      ogy for teaching the skills needed for             ing ergonomics (tool design), robotics (ro-
      successful multidisciplinary research              bot hand design), and anthropology
      projects. The education plan consists of           (research in human hand evolution and
      David’s course, ‘Interdisciplinary Scien-          tool use).
      tific Visualization,’ and a research group;         This summer she was invited to attend
      both are aimed at undergraduates and               the first Asia-Pacific Advanced Studies
      graduate students alike. The research ef-          Institute, with the theme “New Frontiers
      fort includes development of computa-              of Intelligent Robotics,” in Tokyo, Japan.
      tional tools for capturing geometry,               She is on junior sabbatical leave this year,
      representing it within the computer, and           and will be spending much of the year at
      using those representations for specific            Carnegie Mellon University in Pitts-
      applications in archaeology and biological         burgh. She and her students will be using
      modeling.                                          the motion-capture lab at CMU and work-
                                                         ing with Jessica Hodgins’ group on prob-
                         www                             lems related to realistic animation of
                                                         human motion and control of robots from
                                                         motion-capture data.
      Nancy Pollard. Nancy was recently
      awarded an NSF CAREER grant on
      “Quantifying Humanlike Enveloping                                     ww w
      Grasps.” Nancy’s work under the grant
      will focus on quantifying humanlike en-
      veloping grasps with the goal of creating
                                                         John Savage. John chaired the Inau-
                                                         gural Faculty Program Committee that
      credible hand use for digital characters.
                                                         assembled 20 exciting panels reporting on
      Producing realistic digital humans has
                                                         faculty research at President Ruth Sim-
      been called the last frontier in the march
      toward graphical realism, and Nancy be-            mons’ inauguration this October. He is
      lieves that the last frontier in creating          currently Chair of the Faculty and is now
      digital humans is generating believable            in his second year as an elected officer of
      hand motion. In pursuit of this goal, she          the Faculty. As of January 1, John will be
      proposes a tendon-based quality measure            withdrawing as Director of the Industrial
                                                         Partners Program to be replaced by
      for humanlike enveloping grasps, and she
                                                         Michael Black; however, he will work
      plans to evaluate this quality measure (1)
                                                         with Michael over the next year to bring
      for ability to discriminate between
                                                         him up to speed.
      grasps, (2) as a predictor of grasp forces,
      and (3) for use in modeling grasp acquisi-
      tion. Because of the strong emphasis on
      human anatomy, this research has the po-                             www
      tential for additional impact outside
      graphics and animation in areas includ-            Eli Upfal. Eli was on the program com-
                                                         mittee of the 13th International Sympo-
                                                         sium on Fundamentals of Computation
                                                         Theory in Riga, Latvia, and the VIII In-
                                                         ternational Colloquium on Structural In-
                                                         formation and Communication Com-
                                                         plexity in Spain. He participated in a
                                                         Dagstuhl meeting on “Design and Analy-
                                                         sis of Randomized and Approximation Al-
                                                         gorithms” in June and was an invited
                                                         speaker at a workshop in honor of Allan
                                                         Borodin’s 60th birthday at the University
                                                         of Toronto, also in June.

 The 40 TeachScheme! workshop participants
gather for a group shot outside the CIT building

                                                conduit! 26
      David Laidlaw, John Hughes and                        Andy van Dam. Andy participated in
      Andy van Dam. Brown presented four                    two 60th birthday celebrations: On May
      papers at this year’s Symposium on Inter-             29 he gave the keynote address “User In-
      active 3D Graphics (John Hughes was the               terfaces: Disappearing, Dissolving, and
      program co-chair with Carlo Sequin of                 Evolving” at the Celebration Colloquium
      U.C. Berkeley, and Science and Technol-               for Prof. José Encarñaçao, founder and di-
      ogy Center collaborator Mary Whitton                  rector of Fraunhofer’s Graphics Research
      was the general chair). Joe LaViola (Ph.D.            Institute and a faculty member at the
      student of Andy van Dam) presented                    Technical University of Darmstadt. In Oc-
      work that he did with Daniel Acevedo Fe-              tober his paper “Reflections on Next-Gen-
      liz and Daniel Keefe (Ph.D. students of               eration Educational Software” was pub-
      David Laidlaw) and Robert Zeleznik                    lished in a book honoring the career of
      (graphics staff) presented work on                    Prof. Bernard Levrat, computer science
      “Hands-Free Multi-Scale Navigation in                 professor at the University of Geneva and
      Virtual Environments.” Then Dan Keefe                 president of the Swiss Virtual Campus. In
      presented a paper on “Cave Painting,” a               addition, he continued his own celebra-
      joint project with Daniel Acevedo Feliz,              tion of time by hiking the heights and the
      Tomer Moscovich (Ph.D. student of John                depths of the earth—the Alps and the
      Hughes), David Laidlaw and Joe LaViola.               Grand Canyon.
      In Cave Painting, an artist uses virtual
      brushes and paints to create an artwork
      that lives in the 3D space of the four-wall
                                                                            ww w
      Michael Kowalski (graphics staff) pre-                Stan Zdonik. Stan was co-presenter at
      sented a paper on “User-Guided Composi-               the 27th Very Large Database Conference
      tion for Art-Based Rendering,” joint work             (VLDB) in Rome of a half-day tutorial on
      with John Hughes, Cynthia Beth Rubin                  Data Management for Pervasive Comput-
      (of RISD) and Jun Ohya of ATR Research                ing.
      in Japan.
      Finally, Takeo Igarashi, a postdoc in the                              ww w
      graphics group, presented his work with
      Dennis Cosgrove of CMU on “Adaptive
      Unwrapping for Interactive Texture
      Painting” —work inspired by a desire to
      paint colors onto his now-famous “Teddy”

                                                            systems are emerging as a new area of in-
FIVE CS FACULTY AWARDED                                     formation technology research. The physi-
                                                            cal structure and adaptability of the
     NSF’s ITR GRANTS                                       human brain make these biologically em-
                                                            bedded systems quite different from com-
      The National Science Foundation has an-               putational systems typically studied in
      nounced 309 awards designed to preserve               Computer Science.
      America’s position as the world leader of             Fundamentally, biologically embedded
      computer science and its applications.                systems must make inferences about the
      CS’s five newly funded projects were se-               behavior of a biological system based on
      lected from over 2000 competitive pro-                measurements of neural activity that are
      posals.                                               indirect, ambiguous, and uncertain. More-
                                                            over, these systems must adapt to short-
      Michael Black’s grant is for his work on              and long-term changes in neural activity
      “The Computer Science of Biologi-                     of the brain. These problems are ad-
      cally Embedded Systems.” Biologically                 dressed by a multidisciplinary team in
      embedded systems that directly couple ar-             the context of developing a robot arm that
      tificial computational devices with neural             is controlled by simultaneous recordings

                                                  conduit! 27
from neurons in the motor cortex of a sub-           standard parts of all current speech-rec-
ject. The goal is to model probabilistically         ognition systems. There is now evidence
the behavior of these neurons as a func-             that suggests that finer-grained syntactic
tion of arm motion and then reconstruct              analysis can improve current language
continuous arm trajectories based on the             models. Thus this research will enable a
neural activity. To do so, the project will          wide variety of systems to make better
exploit mathematical and computational               use of language input and thus make
techniques from computer vision, image               these systems more accessible to a diverse
processing, and machine learning.                    user pool.

This work will enhance scientific knowl-              Eli Upfal has been awarded a five-year
edge about how to design and build new               grant for his work with Harvard’s Michael
types of hybrid human/computer systems,              Mitzenmacher on “Algorithmic Issues
will explore new devices to assist the se-           in Large-Scale Dynamic Networks.”
verely disabled, will address the informa-           They will develop a theoretically well-
tion technology questions raised by these            founded framework for the design and
biologically embedded systems, and will              analysis of algorithms for large-scale dy-
contribute to the understanding of neural            namic networks, in particular, for the Web
coding.                                              and related dynamic networks, such as
                                                     the underlying Internet topology and In-
Eugene Charniak’s ITR grant is for                   ternet-based peer-to-peer ad hoc net-
work on “Learning Syntactic/Seman-                   works. We plan to develop rigorous math-
tic Information for Parsing.” The pro-               ematical models that capture key charac-
ject is so named because the structural in-          teristics and can make reliable predic-
formation to be learned often falls at the           tions about features such as connectivity,
boundary between syntax and semantics.               information content, and dynamic of these
For example, is the fact that “Fred” is typ-         networks. We plan to apply this frame-
ically a person’s first name a syntactic or           work to test existing algorithms and con-
semantic fact? Does the fact that the                struct improved new algorithms.
“New York Stock Exchange” has as part of
its name the location “New York” fall un-            The main benefits of developing the math-
der syntax or semantics? What about the              ematical models of the Web structure and
similarity between the expression “[to]              dynamics will be the improved theoretical
market useless items” and “the market                foundation for the design, analysis and
for useless items”? These are some of the            testing of algorithms that operate in the
topics that come up in this research.                Web environment. The tangible results of
                                                     this work will be models that can be sub-
As for the “for parsing” portion of the ti-          jected to experimental verification, analy-
tle, the intention is to learn the above             ses of algorithms based upon these
kinds of information in a form that cur-             models, new algorithms that benefit from
rent statistical parsers can use so that             these analyses, and, finally, proof-of-con-
they can output more finely structured                cept demonstrations and experimental
parses. However, this is not meant to sug-           evaluations of such algorithms.
gest that parsing is the sole use for this
sort of information—exactly the opposite             Pascal Van Hentenryck has been
is the case. For example, more and more              awarded a large ITR grant, over four
systems for automatically extracting in-             years, for work in conjunction with Eli
formation from free text use coreference             Upfal and researchers at MIT and Geor-
detection and “named-entity recognition”             gia Tech on “Stochastic combinatorial
(e.g., recognizing that “New York” is a lo-          optimization.” There are many real opti-
cation but “New York Stock Exchange” is              mization problems for which no solutions
an organization). There is evidence to               are known. For instance: a snowstorm is
suggest that both coreference and named-             approaching Chicago and United Airlines,
entity recognition can be improved with              at its Operations Center, must plan how
the finer level of analysis to be made pos-           to cancel and reroute its flights. Although
sible by this research. Or again, “lan-              substantial information on weather fore-
guage models” (programs that assign a                casts, plane and crew status, passenger
probability to strings in a language) are            itineraries, and hotels is available elec-

                                           conduit! 28
tronically, this information is not ex-                mersive 3D virtual environment, much as
ploited in a scientific way. Instead, hu-               if they were co-located in a shared physi-
mans make ad-hoc decisions based only                  cal space. Immersive “time machines” will
on their experience. Why is this the case,             add a further important dimension, that
especially when the airline is a very so-              of recording experiences in which a view-
phisticated user of optimization for plan-             er, immersed in a 3D reconstruction, can
ning? Because the application is a large-              literally walk through the scene or move
scale stochastic combinatorial optimiza-               backward and forward in time. This work
tion problem for which no known algo-                  will focus on a societally important and
rithm produces good solutions in reason-               technologically challenging driving appli-
able time. Our fundamental research in                 cation, teaching surgical management of
this relatively unexplored area will have              difficult, potentially lethal injuries.
two complementary thrusts: (1) exploiting
the orthogonal strengths of constraint                 They will develop a new paradigm for
and mathematical programming to tackle                 teaching surgical procedures: immersive
the hard combinatorial problems arising                electronic books that in effect blend a
in stochastic optimization (e.g., multi-               time machine with 3D hypermedia. Their
stage or Monte Carlo approaches) and (2)               goal is to allow surgeons to witness and
studying stochastic combinatorial sub-                 explore (in time and space) a past surgical
structures that are amenable to efficient               procedure as if they were there, with the
solutions or approximations.                           added benefit of instruction from the orig-
Andy van Dam’s three-year grant,                       inal surgeon or another instructor as well
“Electronic books for the tele-immer-                  as integrated 3D illustrations, annota-
sion age: a new paradigm for teach-                    tions, and relevant medical metadata.
ing surgical procedures,” will directly                The trainee should be able freely and nat-
involve researchers from the University of             urally to walk around a life-sized, high-fi-
North Carolina, Chapel Hill, as well as                delity 3D graphical reconstruction of the
Brown and indirectly researchers from                  original time-varying events, pausing or
UPenn, working with those from UNC on                  stepping forward and backward in time to
realtime 3D model acquisition and recon-               satisfy curiosity or allay confusion. The
struction. Their work on tele-immersion                researchers will bring together experts in
will provide a dramatic new medium for                 the respective disciplines and leverage
groups of people remote from one another               their prior work in tele-immersion in or-
to work and share experiences in an im-                der to achieve these goals.

                  Tele-immersion technology will let surgeons-in-training
                move naturally within a life-sized, high-fidelity, 3D graphical
                reconstruction of the surgery, pausing or stepping forward
                         and backward in time to assist in learning

                                             conduit! 29
           Takeo Igarashi, a postdoc in the Graphics Group here working with John Hughes, hit the worldwide news October
           18. The BBC News Online, no less (, picked up
           a system he’s to present at the ACM UIST conference in Orlando in November. What’s more, (http://slash-
 , which calls itself “News for Nerds” and is read, John Bazik
                                  says, by “millions of geeks like me,” just grabbed the story and featured it; it’s now exciting lots of
                                  commentary, both insightful and amusing, in slashdot’s feedback section.
                                 Takeo’s innovative idea is to use simple human sounds, like grunts and sighs, for controlling com-
                                 puters. Conventional voice-recognition software is still not accurate or efficient enough for general
                                 use in an interface. Takeo’s system works by measuring the pitch and duration of grunt-like sounds
                                 like “ah” and “umm.” Thus, Takeo suggests, in scrolling through a document on the Internet you
                                 could say “move down, ahhhh”: the document would continue scrolling as long as the sound contin-
                                 ued. You could increase the scrolling speed by raising the pitch of your voice, and stop scrolling al-
                                 together by stopping speaking. It’s much easier to Undo a command with a quick “uh oh” than with
                                                             a mouse!
                                                              Said Takeo of his system, “I personally do not think this is useful in of-
                                                              fice environments—it is really annoying for office mates! I basically de-
                                                              signed it for isolated situations such as controlling computers while
                                                              driving a car or rock climbing. I also think the technique is ideal for en-
                                                              tertainment applications. I implemented a simple video game using the
                                                              technique, and children love it.” You can find Takeo’s paper and demo
                                                              video at his homepage,

                                                                                   to learn that he used Drew McDermott’s
   CHARNIAK UNPLUGGED                                                              and my textbook) but was not happy with
                                                                                   the result, which did not sound very much
                                                                                   like his music.
                 I was recently invited to participate in a
                 panel discussion with the title “Is it Art                        He then did what any good scientist does
                 Yet?” The main panelist was a composer,                           when confronted by a too-difficult prob-
                 David Cope, who has written a computer                            lem—he found a simpler one. He figured
                 program that composes music in styles                             that while he was not sure what proper-
                 ranging from Bach to Stravinsky that has                          ties made his music sound like “him”, he
                 fooled experts. The question for the panel                        did know the properties that made Bach
                 was captured by the title.                                        sound like Bach and Mozart like Mozart.
                 I would say that my performance on the                            The program he eventually came up with,
                 panel was forgettable if I could only re-                         however, was not restricted to any partic-
                 member what I said. Actually, most of the                         ular composer. Rather, one gives it a data-
Eugene           other panelists were only so-so as well.                          base of musical works, and the program
Charniak         But fortunately our invited guest, Profes-                        tries to discover the similarities and dis-
                 sor Cope, was great. He is a funny, articu-                       similarities and then use them to create
                 late guy, with a combination of down-to-                          more music with the same features.
                 earth personality and controversial opin-
                                                                                   Eventually Professor Cope finished the
                 ions that make him ideal for a panel dis-
                                                                                   commission. The work, he says, is about
                 cussion. But what I found most intriguing
                                                                                   60% him, 40% computer. However, from
                 was the personal story behind the cre-
                                                                                   his point of view all of the work is his. In-
                 ation of this program.
                                                                                   deed, these days it can be hard to tell
                 It seems that in the early eighties Prof.                         where one leaves off and the other begins.
                 Cope received a commission for an opera.                          Now when he composes he gives the pro-
                 He received (and spent) the advance, but                          gram a database of his own work and af-
                 unfortunately suffered a major composer’s                         ter composing a bit asks the machine to
                 block. Finally, after a long period of time,                      suggest, say, what the next two measures
                 he hit upon the idea of creating a com-                           should be. If he does not like the sugges-
                 puter program to compose for him. So he                           tion he can keep asking for other alterna-
                 taught himself Lisp and AI (I was pleased                         tives.

                                                                         conduit! 30
   His major work at the moment is not in                dents (female) were involved in a prosti-
   his own modern idiom but rather an op-                tution ring. “Ivy League Madam!” or some
   era about the life of Mahler, written (with           such. Brown immediately went into full-
   the help of his program) in the style of              court damage control, with multiple press
   Mahler. Mahler, besides being a composer,             conferences all saying that what students
   was an active conductor, and although he              did on their own time was beyond our con-
   frequently conducted operas, he never                 trol, that this was not exactly what we
   wrote one. David Cope intends to correct              thought the life of the mind was all about,
   this historical mistake.                              and that we did prefer students without
                                                         criminal records.
   I teach CS002, our department’s com-
   puter literacy course. During the semes-              At the time I thought that Brown was
   ter the students, among other things,                 overreacting, but now I am not so sure. A
   learn the basics of HTML, spreadsheets,               key fact in the above is that it occurred in
   etc. For the final project in the course the           early spring—the time when students
   students have the option of a more exten-             (and their parents) are deciding where to
   sive project in any of the above, or learn-           go to college. I have reconsidered because
   ing a new software package. One of last               this last spring I was one such parent. In
   year’s students briefly put her project on             particular, one of the schools my son was
   the web. She and a friend bought the web              considering was Hamilton, a small, selec-
   address where                  tive, liberal arts college in upstate New
   one was able to fill out a form specifying             York. In early spring the New York Times
   the sex, skin color, eye color, price one is          published an article about cloning hu-
   willing to pay, etc. However, all you got for         mans prominent in which was one Profes-
   your pains is a notice that it is a joke. If          sor Boisselier, identified as a professor in
   she didn’t mind getting the FBI after her,            the chemistry department at Hamilton.
   it would have been a better joke if she’d             Worse, the organization whose cloning
   asked for MasterCard or Visa.                         laboratory she runs, the Raelians, is
   About 15 years ago or so, during the early            headed by a fellow named Rael, who was
   spring, the tabloids had a field day when              abducted by aliens and seduced by female
   it was discovered that several Brown stu-             alien robots, and who believes on the ba-
                                                         sis of this that the destiny of the human
                                                         race is to make itself perfect and immor-
                                                         tal through the use of science.
                                                         After some investigation it turned out she
                                                         was a visiting faculty member, she re-
                                                         signed from her position, my son decided
                                                         to go to Colby, and he thought the whole
                                                         thing was a hoot. But I mentioned this to
                                                         a neighbor who is the spokesperson for
                                                         the Rhode Island School of Design. She
                                                         responded that a few years ago, again in
                                                         the spring, several RISD students were
                                                         arrested by the Boston police after nearly
                                                         finishing painting an entrance to an
                                                         MBTA station pink. A picture appeared
                                                         on the front page of the Boston Globe with
                                                         the headline ‘Caught Pink-Handed’.
                                                         When the students were asked what they
                                                         were doing they responded that this was
                                                         their ‘site-specific’ art work for one of
                                                         their RISD courses. Naturally, this fell in
                                                         my neighbor’s lap. A reporter with whom
No, Suzi Howe did not dress to match the                 she had dealt before called to ask for
 tablecloths on IPP Symposium day, but                   RISD’s position. She responded that RISD
 everyone who stepped off the elevator                   did not have a position, but would he take
            thought she had!                             a comment off the record? Off the record,
                                                         RISD’s position was ‘What shade of pink?’

                                               conduit! 31
                               Of the more than 5,000 feared dead as a result of the terror-
                               ist attacks on September 11, at least six were Brown
                               As far as we know, none of them had connections with this
                               department. Members of the Brown University community
                               paused during three events in the October 12-14 celebration
                               of President Ruth Simmons’ inauguration to remember and
                               honor these alumni.
                               For details and a list of those lost, see:

                                   A publication of
                           The Computer Science Department
                                   Brown University
                                    Inquiries to: conduit!
                             Department of Computer Science
                                Box 1910, Brown University
                                   Providence, RI 02912
                                     FAX: 401-863-7657
                                  PHONE: 401-863-7610
                                EMAIL:                             Katrina Avery
                                                                Suzi Howe                              Jeff Coady
                                                              Editor-in-Chief           Editor      Technical Support

Department of Computer Science
          Brown University                                                                           NON-PROFIT
Box 1910, Providence, RI 02912, USA
                                                                                                     U.S. Postage
                                                                                                    Providence, RI
                                                                                                      Permit #202

Printed on recyled paper                                                                     Address changes welcomed

                                                                      conduit! 32
                                   A publication of
                           The Computer Science Department
                                   Brown University
                                    Inquiries to: conduit!
                             Department of Computer Science
                                Box 1910, Brown University
                                   Providence, RI 02912
                                     FAX: 401-863-7657
                                  PHONE: 401-863-7610
                                EMAIL:                             Katrina Avery
                                                                Suzi Howe                              Jeff Coady
                                                              Editor-in-Chief           Editor      Technical Support

Department of Computer Science
          Brown University                                                                           NON-PROFIT
Box 1910, Providence, RI 02912, USA
                                                                                                     U.S. Postage
                                                                                                    Providence, RI
                                                                                                      Permit #202

Printed on recyled paper                                                                     Address changes welcomed

                                                                      conduit! 33

To top