Docstoc

RANDOMNESS_ PROBABILITY and COMPUTING “how can a string of random

Document Sample
RANDOMNESS_ PROBABILITY and COMPUTING “how can a string of random Powered By Docstoc
					Volume 8, Number 2
                      conduit!
                      condu t     Department of Computer Science
                                         Brown University
                                                                                                       Fall, 1999



                                                                          in the Ethernet card, uses random num-
     RANDOMNESS, PROBABILITY                                              bers to decide the exact timing by which
                                                                          it tries to access the Ethernetcommuni-
         and COMPUTING                                                    cation medium. Other commonly used
                                                                          applications of randomized algorithms
                              “God does not play dice with the
                                                                          are Monte-Carlo simulations and pri-
                              universe” was Albert Einstein’s
                                                                          mality testing for crypto-graphy applica-
                              famous objection to modern
                                                                          tions. In these and many other
                              quantum-mechanics theory. This
                                                                          important applications, randomized al-
                              and similar objections reflect the
                                                                          gorithms are significantly more efficient
                              human conceptual difficulty in
                                                                          than the best known deter-ministic solu-
                              grasping a world that is not gov-
                                                                          tions. Furthermore, in most cases the
                              erned by fully deterministic
                                                                          randomized algorithms are also simpler
                              laws. Nevertheless, research in             and easier to program.
                              the past century has demon-
                              strated the dominance of random             A randomized program can be viewed as
                              behavior and statistical laws in            a conventional (deterministic) program
                              almost any field of science,                 that reads two streams of inputs: one is
                              ranging from sub-particle phys-             the actual input of the program, the
                              ics to free market economics.               other is a sequence of randomly gener-
                              Computer science is no excep-               ated numbers. This view raises an in-
       Eli Upfal
                     tion: from the highly theo-
                     retical notion of probabilistic  “how can a string of random
                     theorem proving to the very
                     practical design of PC Ether-    numbers (perfect noise) that
                     net cards, randomness and
                     probabilistic methods play a
                                                     adds no information to the pro-
                     dominant role.                  gram actually improve the per-
                     Roughly speaking, random-         formance of its execution?”
                     ness comes up in two aspects
                     of the study of algorithms: randomized                triguing philosophical question: how can
                     algorithms and probabilistic analysis of              a string of random numbers (perfect
                     algorithms. Randomized algorithms are
                                                                           noise) that adds no information to the
                     algorithms that make random choices
                                                                           program actually improve the perfor-
                     during their execution. In practice, a
                     randomized program uses values gener-                 mance of its execution? Theoretical com-
                     ated by a random number generator to                  puter science still does not have a full
                     decide the next step at several branches              explanation of that phenomenon, though
                     of its execution. For example, the                    more and more applications take advan-
                     Ethernet access protocol, implemented                 tage of it.


                                            Brown University, Box 1910, Providence, RI 02912, USA
Another algorithm-related application of                  mance of dynamic processes, in partic-
probability theory is in analyzing the av-                ular through modeling the dynamic
erage-case performance of algorithms.                     process as an infinite stochastic process.
Algorithm and complexity theory tries to
classify computation problems ac-                         A key feature of dynamic algorithms is
cording to their computational com-                       the need to make decisions ‘on-line’
plexity, in particular distinguishing                     without information about future re-
between easy and hard to solve prob-                      quests. An obvious example is cache
lems. For example, complexity theory                      maintenance protocols. A cache is a fast
shows that the famous traveling                           memory that serves as a buffer between
salesman problem is NP-hard. Thus it is                   the processor and a slower but larger
very unlikely that it has a solution sig-                 memory module. Since a cache is typi-
nificantly faster than enumerating all                     cally small, the program often requests
possible routes—a task that is exponen-                   pages that are not in the cache. When a
tial in the number of cities and thus not                 new page is brought to the cache, some
feasible for a large number of cities.                    other page must be returned to the main
                                                          memory. Accessing that page later will
An embarrassing phenomenon for the                        be expensive, and thus the algorithm
classical ‘worst-case’ complexity theory                  tries to evict from the cache pages that
is that problems classified as hard to                     are less likely to be requested in the fu-
compute by the theory are often easy to                   ture. The algorithm makes these deci-
solve in practice. Probabilistic analysis                 sions, of course, without actually know-
gives a theoretical explanation of that                   ing the future sequence of requests.
phenomenon, namely that these prob-
lems are hard to solve on some small set                  As in the case of static algorithms, ran-
of pathological hard inputs but are actu-                 domness is introduced into dynamic
ally easy to solve on most inputs, partic-                computation through the algorithm, the
ularly those that come up in real-life                    input or both. Many interesting dynamic
applications—scheduling, the traveling                    protocols, such as the Ethernet access
salesman problem, packing and covering                    protocol mentioned before, are random.
are just a few examples.                                  An execution of a dynamic random algo-
                                                          rithm, even on a fixed input sequence,
Dynamic Algorithms                                        defines an infinite stochastic process in
The focus of my research is the applica-                  which a state at a given step depends on
tion of probability theory in computer                    the history of the process. Analysis of
science. In particular, I am currently in-                such a process requires a different ap-
terested in studying dynamic
algorithms     through
chastic an-‘alysis. Research
                            sto-   “Worst-case analysis rarely
in theoretical computer sci-       gives interesting insight into
ence has focused mainly on
static computation problems,
                                  the actual performance of a
where the input is known be-          dynamic algorithm”
fore the start of the computa-
tion and the goal is to minimize the                      proach and different tools than those
number of steps till termination with a                   used in analyzing the finite execution of
correct output. However, many impor-                      a randomized static computation.
tant processes in today’s computing are
dynamic or interactive processes, where-                  Stochastic Analysis of Dynamic
by input is continuously injected to the                  Algorithms
system and the algorithm is measured                      Worst-case analysis rarely gives inter-
by its long-term steady-state perfor-                     esting insight into the actual perfor-
mance. Examples of dynamic processes                      mance of a dynamic algorithm. A worst-
include communication protocols, mem-                     case adversary can generate extremely
ory-management tools, and time-sharing                    hard sequences of requests, and the per-
policies. The goal is to develop new tools                formance of the algorithm on these
for designing and analyzing the perfor-                   pathological cases does not accurately


                                             conduit! 2
represent its efficiency. To offset the ef-
fect of rare cases, it is useful to analyze
the performance of dynamic systems
under some stochastic assumptions on
the stream of inputs. Such assumptions
are more realistic in dynamic settings,
especially when requests are originated
by a number of independent processors,
than in static analysis. The stochastic
process that controls the stream of re-
quests might be stationary, periodic, or
even bursty. The goal is to obtain results
that are valid under the weakest set of
assumptions. The advantages and prac-
ticality of this approach have been well
demonstrated by the achievements of
queuing theory. Our goal is to apply sim-
ilar techniques to dynamic computer
                                                                    Gopal Pandurangan
processes that do not fit the queuing
theory settings.                                           due to events such as link and buffer
Stochastic analysis of dynamic processes                   overflow, packet loss, jitter, etc.
builds on the rich theory of stochastic                    Our new protocol employs an efficient
processes, in particular queuing theory,                   Monte-Carlo method for estimating the
and the theory of stationary processes.                    failure probability of a network. The
However, in many cases new tools are                       tight estimate allows the protocol to uti-
needed to address the specific problems                     lize the network bandwidth fully with-
posed by computer-related processes                        out violating the QoS requirements. The
that are discrete and involve compli-                      new method is particularly useful in a
cated dependency conditions.                               dynamic setting in which communica-
                                                           tion requests are dynamically added to
Recent Work                                                and eliminated from the system. The
PhD candidate Gopal Pandurangan and                        amortized cost in our solution of up-
I are studying an improved protocol for                    dating the estimate after each change is
admission control in fast communication                    proportional to the fraction of links in-
networks. Modern communication proto-                      volved in the change rather than to the
cols such as ATM (asynchronous                             total number of links in the network. We
transfer mode) achieve high utilization                    are currently running an extensive sim-
of channel bandwidth by multiplexing                       ulation project to evaluate the perfor-
communication streams with different                       mance of the new technique under
flow characteristics into one communica-                    various network and load settings.
tion channel. Requests for communica-
tion are submitted to the network                          Course
management protocol with some statis-
tical characterization of the required                     The research area combines two disci-
communication. The network (flow)                           plines, algorithm theory and probability
management protocol uses this informa-                     theory. My new course ‘CS155: Probabi-
tion to statistically multiplex as many                    listic Methods in Computer Science’ ex-
communication requests as possible                         poses students to the interplay between
while maintaining global network per-                      these two areas. They study basic proba-
formance. Next-generation communica-                       bility theory, in particular discrete prob-
tion networks are expected to provide                      ability theory (which is more relevant to
QoS (quality of service) guarantees                        computer science applications) and then
when satisfying communication re-                          apply this theory to design and analysis
quests. In particular, QoS protocol is ex-                 of randomized computer algorithms for a
pected to limit to a pre-specified value                    variety of applications.
the probability of communication failure


                                              conduit! 3
                        SAVAGE HONORED ON HIS 60th
                   On September 21, in honor of John Savage’s 60th birthday, Franco Preparata hosted a
                   one-day technical forum in the department on ‘‘Algorithmic Research.’’ The event recog-
                   nized, in its subject and character, John’s constant devotion to productive and well-
                   regarded academic research. Colleagues and former doctoral students dedicated to the
                   honoree some of their more recent work. The department participated actively in the
                   event, recognizing John’s many contributions as a co-founder of our department and his
                   assiduous stewardship as chairman and excellent citizen. After initial remarks by Tom
                                                                         Dean and Andy van Dam, Franco
                                                                         kicked off the day with a talk on
                                                                         “Reconstructing a sequence from its
                                                                         samples: DNA sequencing at the
                                                                         information-theory bound,” Charles
                                                                         Fiduccia (PhD ’73, Division of Engi-
                                                                         neering) spoke on “Optimal mono-
                                                                         tonic search,” Eli Upfal on
                                                                         “Reducing    network       congestion
                                                                         through balanced allocation,” David
                                                                         Carlson (PhD ’80) on “New insights
                                                                         into two old algorithms: the Fast
                                                                         Fourier Transform and metropolis
                                                                         Monte-Carlo method,” and Roberto
                                                                         Tamassia on “Graph drawing and
                                                                         information visualization.” The
                                                                         intense technical program had a
                                                                         relaxing break during the buffet
                                                                         luncheon and was capped by a gra-
                                                                         cious departmental reception (with
Back, l to r: David Carlson, Roberto Tamassia, Eli Upfal. Front, l to r: birthday cake, of course).
         Charles Fiduccia, John Savage, Franco Preparata



                                                                            IJCAI is the premier international con-
           INSIDE IJCAI ’99                                                 ference in AI and the invitation from
                                                                            then-president Ray Perrault was very
                                                                            flattering. Stockholm was one of my fa-
                                Nearly four years ago, at                   vorite cities, I enjoyed working with Ray
                                the 1995 IJCAI (Interna-                    and the other trustees, and I thought
                                tional Joint Conference on                  that my experience running other con-
                                Artificial Intelligence) con-                ferences (in addition to smaller confer-
                                ference in Montreal, I was                  ences and workshops, I was program
                                asked to be program chair                   chair for NCAI in 1991—the National
                                for the 1999 IJCAI in                       Conference on AI, often called AAAI for
                                Stockholm and also to                       the North American professional associ-
                                serve on the board of                       ation that runs it) would make this new
                                trustees for IJCAI Inc., the                task relatively straightforward. I naively
                                governing board that runs                   agreed to serve.
                                the IJCAI conferences,
                                oversees a trust fund, de-                  I can’t begin to say how unprepared I
                                termines conference ven-                    was for the experience. NCAI is man-
                                ues, awards prizes, and                     aged by a well-run professional organiza-
                                generally plays an impor-                   tion with relatively deep pockets. Much
                                tant role in the field                       of what was necessary in running AAAI-
                                through the various activi-                 91 was handled without my being aware
                                ties associated with the bi-                of it by Carol Hamilton, the Executive
Tom Dean in Stockholm           ennial IJCAI conferences.                   Director for AAAI, and her excellent

                                                               conduit! 4
       staff. I was oblivious of many of the de-                 with the conference, and Gigina, Anita
       tails of running a large conference. Even                 and I had to direct the troops.
       though IJCAI is a larger conference than
                                                                 Running an international conference is
       NCAI and much more complicated in its                     fraught with all sorts of politics. You
       international focus, most IJCAI work is                   have to make sure that all the appro-
       done by the trustees and numerous vol-                    priate nationalities and professional so-
       unteers from the academic and research                    cieties are represented. Even though the
       communities. The trustees turn over reg-                  conference was to be held in Stockholm,
       ularly, so corporate memory is short. For                 it was supposed to be hosted by all the
       the most part, each IJCAI is invented                     Scandinavian countries. Registration
       anew, with little help from past experi-                  typically doesn’t cover the cost of run-
       ence.                                                     ning a conference, and so to break even
                                                                 we needed to get sponsors willing to help
       Each IJCAI has a program chair in                         out financially but not require their cor-
       charge of the technical program, a gen-                   porate logos to be tattooed on the fore-
       eral chair in charge of the whole process,                heads of the attendees.
       a conference arrangements chair, and nu-
       merous other local and special program                    Much of my effort was directed at put-
       chairs. I worked most closely with Prof.                  ting together the program committee,
                                                                 finding chairs for workshops and tuto-
       Luigia Carlucci (Gigina) Aiello, the gen-
                                                                 rials, and choosing invited speakers and
       eral chair and a professor at Università
                                                                 cajoling them to attend even though we
       di Roma “La Sapienza,” and Prof. Anita
                                                                 could only cover a portion of their overall
       Kollerbaur, the conference arrangements                   expenses. I enlisted the help of 37 top re-
       chair and a professor at Stockholm Uni-                   searchers to serve on the program com-
       versity and Royal Institute of Technology.                mittee, and these folks in turn enlisted
       Gigina and Anita were wonderful to work                   the help of nearly 400 reviewers to
       with, but our job was incredibly frus-                    handle the almost 800 submissions and
       trating and a bit unnerving as we tried to                over two thousand reviews. I won’t bore
       figure out how to run a conference with a                  you with the details of coordinating this
       more than million-dollar budget (and no                   effort and coping with inevitable glitches
       small amount of financial risk) starting                   and embarrassments that occur when
       pretty much from first principles.                         dealing with so many people and trying
                                                                 to adhere to a strict deadline determined
 “Running an international                                       by printers, international shipping rules,
                                                                 and of course the actual date of the con-
 conference is fraught with                                      ference at which the final proceedings
                                                                 are to be handed out to attendees.
all sorts of politics. You have
                                                                 By the end of last February, we had se-
  to make sure that all the                                      lected 195 papers for presentation, 14 in-
  appropriate nationalities                                      vited speakers were lined up, 29 work-
                                                                 shops and 20 tutorials were on the pro-
       are represented”                                          gram, and we had made the selections
                                                                 for the Computers & Thought and Re-
       Well, we didn’t really start from first                    search Excellence awards. The locations
       principles; Anita knew a lot about plan-                  for the conference, the receptions, the
       ning large projects (part of her academic                 banquet and other events were settled.
       expertise) and she drew upon the skills of                Given the chaos of reinventing IJCAI
       the administrative and technical staff at                 and the fact that none of us were exactly
       her university and hired a professional                   sure who was responsible for what, I was
                                                                 relieved that Gigina, Anita and I were
       conference organizer to help with various
                                                                 still speaking with one another. After
       aspects of the planning and execution. I
                                                                 what seemed like two years of steady
       arranged for Carol Hamilton and AAAI to
                                                                 work, I thought my job was over.
       help with the extensive correspondence
       and paper-handling associated with the                    Unfortunately, there were some little de-
       technical program. In fact, we had to put                 tails yet to go: putting together the pro-
       together a small army of people to help                   ceedings, developing the content for the

                                                    conduit! 5
brochure, providing a schedule for the                    people took advantage of the light to
conference program, creating paper ses-                   wander in the old town (Gamla Stan),
sions and assigning them to rooms,                        enjoy a walk around Skeppsholmen, or
finding reliable session chairs and people                 take in the rides and other amusements
to introduce invited speakers (92 people),                in Djurgarden. The trustees plotted and
and a myriad of other details. It seemed                  planned and were rewarded for their ef-
that I would never get out from under                     forts by a wonderful dinner in a restau-
the yoke of IJCAI.                                        rant located in one of the royal game
                                                          parks. There were more courses than I
My wife Jo and I arrived in Stockholm on
                                                          could count and fabulous wines that I’ll
July 30, several days before the official
                                                          not likely sample again any time soon.
beginning of the conference. There were
                                                          One of the trustees knew a little about
tutorials and workshops to check on, a
                                                          wine and the steward took us to visit
robotic soccer competition that was co-lo-
                                                          their cellar; along one wall was a collec-
cated with IJCAI, and trustees’ meetings
                                                          tion that included one bottle from each of
to attend. Indeed, trustees’ meetings con-
                                                          the major houses in the Bordeaux region,
tinued throughout the conference, deal-
                                                          starting from 1900 and extending to the
ing with the previous conference (in
                                                          present. The only missing years were
Nagoya), the current conference, the
                                                          those in which war had eliminated an
next conference (which is to be held in
                                                          entire year’s production.
Seattle and for which the planning was
                                                          I heard very few technical papers at
                                                          IJCAI ’99 and, though I heard parts of al-
                                                          most every invited talk, for one reason or
                                                          another I was never able to sit through
                                                          an entire talk. After the last of the tech-
                                                          nical sessions, I felt completely drained.
                                                          Luckily several of Jo’s family had arrived
                                                          in Stockholm the previous day and so I
                                                          had plenty to distract me from my post-
                                                          partum blues. Jo’s sister Nancy and her
                                                          daughter Lila had flown in from Paris,
                                                          Nancy’s husband had flown in from
                                                          Boston, and Lila’s boyfriend came from
                                                          Rome right after he finished work on
                                                          Friday. We spent a pleasant weekend en-
                                                          joying what the conference’s attendees
                                                          had been experiencing the week before.

  Skeppsholmen                                            In retrospect, it was a grueling four
                                                          years. The final result was very grati-
well under way), deciding on the location                 fying but I don’t know that I would do it
for 2003 (Mexico), and dealing with var-                  again. Delayed gratification is common
ious aspects of planning for 2005, 2007                   in the academic world: you submit a
and beyond.                                               paper or proposal and wait months for a
                                                          response. At this point in my life, how-
The weather was perfect. Stockholm is a                   ever, I think I want my feedback a little
beautiful city and it is great fun to tour                more immediately. Running a confer-
its shops and museums, flit from island                    ence like IJCAI is a bit like building a
to island on its convenient water taxis,                  house: you really can’t enjoy the house
take day trips out to the archipelago, and                until the roof is on, the plumbing work-
relax in its ample parks and gardens.                     ing, the walls painted, and it’s ready to
With the exception of a few rain showers,                 move in; in the interim, there is always
the perfect weather, dry and warm but                     something to do, most of it tedious. Still,
not hot, continued throughout. The con-                   I have to admit that it’s pretty exciting
ference attendees were treated to a boat                  when it all comes together. Today
trip and a banquet on a fortress island.                  someone called me about running an in-
It stayed light till 10 pm or so and most                 ternational conference on ...


                                             conduit! 6
                    Peter and Judith Wegner at the Rehabilitation Hospital of RI in N.
                     Smithfield, where Peter had been undergoing physical therapy.
                            Happily, he is now back at home on the East Side
                                                                  Wegner is now at the Rehabilitation Hos-
                                                                  pital of Rhode Island in North Smith-
PETER WEGNER ON THE MEND                                          field. He returned to the United States
                                                                  Sept. 30, nearly four months after the ac-
        The following article and photograph above are            cident. On June 11, Wegner had been
        by senior news writer Kristin Cole. They                  planning to take a walk in Trafalgar
        appeared in the October 15-21 issue of the                Square, one of his favorite haunts. He
        George Street Journal.
                                                                  had just retired after 30 years of teaching
        The music of Brahms filled the London                      at Brown and was looking forward to a
        hospital room where Peter Wegner, a re-                   summer filled with engagements.
        tired Brown computer science professor,                   His first stop in England was to attend a
        had only recently regained consciousness                  reunion of Jews who, as children, had es-
        after being in a coma for four weeks.                     caped Nazi terror in their homelands
                                                                  when they were accepted by England as
        The bus that struck him as he walked to-
                                                                  political refugees. Wegner had left Vi-
        ward Trafalgar Square a month earlier                     enna as a 6-year-old, one of 10,000 chil-
        had smashed an elbow, broken ribs, and                    dren on the “Kindertransport.” After the
        resulted in a head injury from which doc-                 reunion, he was slated to deliver
        tors had initially predicted only a 5- to                 speeches to professional conferences in
        10-percent chance of survival.                            Portugal and Scotland. To cap the Euro-
                                                                  pean trip, Wegner was to receive Aus-
        When asked whether he liked Brahms,
                                                                  tria’s highest academic award, for his
        Wegner replied, “Yes, especially Boston                   lifetime contribution to the field of com-
        Brahms” —a pun referring to the upper-                    puter science.
        class intellectual Brahmins of 19th-cen-
                                                                  His wife had left Wegner only a short
        tury Boston. For friends and colleagues
                                                                  time before he headed to Trafalgar
        who had given Wegner a chair engraved
                                                                  Square, returning to her sister’s home,
        “to an unrepentant punster” as a retire-                  where they were staying, to help prepare
        ment gift, it was perhaps the surest sign                 dinner. But when the meal was ready
        that he was on the road to recovery. “It                  and Wegner had not arrived home, the
        was reassuring to know it was the same                    family began to worry. They notified po-
        old Peter,” said Judith Wegner, his wife of               lice and learned a short while later that
        43 years. “He beat the odds.”                             Wegner had been taken by helicopter


                                                         conduit! 7
                              from the accident to the Royal London                   In therapy Wegner has strengthened his
                              Hospital. His wallet had not contained in-              ability to walk, read and write. A recent
                              formation for contacting his sister-in-law.             milestone in his recovery was walking
                              But the wallet did contain names of col-                around the perimeter of the hospital
                              leagues, including Brown faculty mem-                   building without the aid of a cane. “Ju-
                              bers Andries van Dam and Thomas                         dith has helped me enormously during
                              Doeppner.                                               this time,” he said. Wegner looks forward
                                                                                      to the day he will return to his office; al-
                              Initial reports about Wegner’s condition                though retired, he still plans to continue
                              were pessimistic, and the couple’s sons,                his research. There are also retirement
                              Mark, Jeremy and Michael, flew from the                  dreams to satisfy, said Judith, such as
                              United States to London to be at the hos-               cruises to Alaska and the Greek Islands.
                              pital. In the weeks that followed, the                  And he has yet to receive the Austrian
                              family updated friends and colleagues on                Medal of Honor for Science and Art,
                              Wegner’s status by e-mail, maintaining a                which the government offered to confer in
                              list that reached across the Atlantic, in-              the hospital. Wegner refused. He wants
                              cluding many at Brown. Additionally, Ju-                to travel to the country to receive the
                              dith notified her husband’s professional                 award, as planned.
                              commitments of the situation.
                                                                                      They are all plans that friends and col-
                              In the hospital room, Judith and her                    leagues thought Wegner would someday
                              sister, Marion Rosenberg, sang familiar                 be able to satisfy after learning that his
                              folk songs to Wegner in English and                     ability to make puns was intact only days
                              German, in which he is fluent. They cele-                after coming out of the coma. “It certainly
                              brated his 67th birthday there, papering                showed the sign that he was recovering,”
                              the walls with cards that wished him well               said Doeppner, associate professor of
                              and often contained puns. When Wegner                   computer science. “He can come up with
                              first came out of the coma, a tube in his                the most amazing puns in zero time.” Ju-
                              throat prevented him from speaking, but                 dith maintains a list of Wegner’s recent
                              it was obvious he could understand what                 puns at their home in Providence, where
                              was being said and recognized his family,               the chair with the engraving “to an unre-
                              said Judith. “That is an unusually long                 pentant punster, indefatigable scholar,
                              coma for someone to recover from. He is a               and generous friend” awaits his return.
                              fighter. He is a determined person and he
                              is working very hard at it now.”


Said Ed Lazowska in his Andyfest kick-off speech, “It’s hard to believe Andy is 60. He looks only 45, same as he did when he was 30!”

  Scenes from last May’s “Andyfest,” a two-day symposium in celebration of




   (Not ZZ Top) Andy and, l to r, Hendrik-Jan Thomassen,                      Guests seated for lunch on campus
   Steve Carmody, Wolfgang Millbrandt and Ken Sloan


                                                                             conduit! 8
                                            VISUALIZING PROGRAM EXECUTION
                      Introduction                                          derstand a particular aspect of the be-
                                                                            havior of a particular system. The third
                      Today’s software systems are complex,                 involves processing the raw trace data so
                      and their behavior, i.e. what happens as              as to obtain the information needed for
                      they execute, is typically even more so.              the visualization. The final problem in-
                      Here one must deal with the interaction               volves actually producing a visualization
                      of large amounts of code, distributed com-            with which the programmer can interact
                      putations, external systems, multiple                 to obtain the desired insights.
                      threads of control, locks, asynchronous
                      events, message traffic, and other compli-             Obtaining the Data
                      cating issues. However complex this be-
                      havior may be, programmers must be                    A lot of data is needed to describe the ex-
                      able to understand what is going on, espe-            ecution of a system. This includes func-
                      cially when the systems start to act in an            tion calls and returns; thread creation,
                      unexpected or erroneous manner.                       destruction, and state changes; the state
                                                                            of locks and other synchronization mech-
                      Our approach to this problem of software              anisms; memory access and paging be-
                      understanding is to use information gath-             havior; local control flow; timing infor-
                      ered as the system executes to give the               mation, both real-time and execution-
                      programmer sophisticated visualizations               time; message passing; files loaded and
   Steven Reiss       describing the execution. The amount of               used; and file and socket utilization. The
                      data that can be collected is vast and can            general rule for software understanding
                      easily be overwhelming. Visualization                 is that you want to collect as much data
                      uses the brain’s sophisticated recognition            as possible since there is some potential
                      abilities to quickly find relevant patterns            problem for which that data will be the
                      in a sea of data in order to make execu-              key to understanding.
                      tion understanding practical.
                                                                            We currently have two separate systems
                      Our research here involves four related               that can gather much of this trace data.
                      but separable problems. The first involves             The first, AARD, works with C or C++
                      obtaining and storing the data. The                   programs. It consists of a package Wolf
                      second involves letting the programmer                that takes executable files and rewrites
                      specify what needs to be visualized to un-            them with instrumentation code, and a




Andy’s 60th birthday, entitled “The Computer, the Academy, and the World.”




Enjoying an outdoor break in front of the new Mac-       President Gee presenting a handsome photograph
 Millan building where most of the talks took place      of the University at the beginning of the banquet at
                                                                               the Biltmore

                                                                   conduit! 9
                         package Vark that uses the raw instru-              The second approach involves under-
                         mentation data to provide data for visual-          standing what information is actually
                         ization. The second system, TMON, uses              needed for a particular visualization and
                         Java’s JVMPI interface to collect trace in-         doing the appropriate analysis and pro-
                         formation on Java programs. It handles              cessing as the raw data is generated so
                         multiple threads of control without                 that only the processed data is stored.
                         adding any additional run time synchroni-           One difficulty here is that we want to let
                         zation and records everything that JVMPI            the programmer define what should be vi-
                         offers.                                             sualized and hence have to compute the
                         Both of these systems generate large                necessary filters and processors dynami-
                         amounts of data. Although it will vary              cally. A second difficulty is that the pro-
                         with different applications, our current            cessing can be quite sophisticated, in-
                         best guess is that they typically produce           volving tracking execution histories,
                         about 1G of raw trace data for every 10             maintaining the execution state, and
                         seconds of raw CPU time. (This is for rela-         looking at the operations of multiple
                         tively high-level information such as               threads without adding additional syn-
                         method calls; with low-level information            chronization that would change how the
                         such as basic block entries or memory ac-           program is behaving.
                         cesses, this goes up by at least an order of
                         magnitude.) Our research here involves
                                                                             Determining What to Visualize
                         managing this trace data so that we can             One conclusion we drew from our past
                         deal with large, long-running systems.              work on software visualization was that
                         We take two different approaches here.              no single or small set of fixed visualiza-
                         The first involves using the increase in             tions will address the wide range of ques-
                         computer capabilities to provide a hard-            tions asked as a programmer attempts to
                         ware-based solution. We are looking at              understand the behavior of a complex
                         doing traces on machines that have sev-             system. Rather than attempting to define
                         eral gigabytes of memory, multiple proces-          such a fixed set and having it not be par-
                         sors, and high-speed connections to large           ticularly useful, we have focused on let-
                         amounts of RAID-based disk storage. We              ting the programmer define what should
                         hope that the combination of such mas-              be visualized and how it should be dis-
                         sive workstations with sophisticated trace          played. By making the foundation of such
                         collection code will let us collect complete        a system rich enough, we hope to address
                         traces with minimal impact on the run-              a wide range of understanding problems
                         ning program.                                       and thus make our system practical.




 Over 300 participants attended Andyfest at venues both on- and off-campus




One of Andy’s first doctoral candidates, Ingrid Carlbom,         With Ed Lazowska ’72 and John Hughes, who
   lands a good one at the roast after the banquet                       orchestrated the entire event


                                                                    conduit! 10
            To this end, we created, as part of our              UNIX prof, and two-level, à la UNIX
            Desert environment, the CACTI front end              gprof), raw call data, call dags (con-
            for defining visualizations. CACTI lets the           verting the dynamic call tree into a dag
            programmer define the data to visualize               by merging common subtrees), memory-
            by graphically defining sets of objects               management data, interval analysis
            using fields that are the various domains             where the execution is broken into inter-
            composing the raw data. It is, in effect, a          vals in which the data is combined, and
            universal-relation-assumption-based vis-             input/output behavior. While all these
            sual query language that can access both             are not available with all the current
            databases describing the static structure            trace packages, we have used each of
            and symbols of the system and the var-               them with real trace information to ad-
            ious dynamic analyses that are available             dress specific behavioral issues.
            from the processed trace data. It is rela-           Our experience here is that more and
            tively easy to use, as we have demon-                better analysis methods are required to
            strated by addressing a wide range of                make better use of the trace data. One
            specific understanding problems in-                   approach we are taking is to look at anal-
            cluding looking for compiler-generated               yses that deal with specific issues such as
            temporaries, abstracting the dynamic call            multiple threads of control (e.g. finding
            graph, exploring memory behavior over                potential deadlocks or race conditions)
            time, looking for patterns in the call stack         and message-passing protocols. Another
            over time, and finding methods and vari-              approach is to use techniques developed
            ables that are never called.                         for data mining to attempt to find pat-
                                                                 terns, both expected and unexpected, in
“Raw trace data by itself is not very                            the trace data. A third approach involves
                                                                 providing a high-level language in which
  useful. Not only is there much too                             the programmer can define the pro-
much information to create mean-                                 cessing to be done. This last approach is
                                                                 interesting in that such a specification
 ingful displays, but the information                            can be used to generate appropriate fil-
                                                                 ters that limit and do some data pro-
 relevant to a particular problem is                             cessing as the trace is generated.
        typically well hidden”                                   Viewing the Result
                                                                 Understanding behavior requires not
            Our continuing research in this area will            only flexibility in the information to be
            attempt to make the CACTI interface                  displayed and the processing to be done
            more intuitive, provide a reasonable front           on the raw trace data, but also in a va-
            end to the large number of potential trace           riety of different visualization strategies.
            analyses, and handle multiple databases              Different visualization techniques are
            with large numbers of potentially over-              appropriate to different types of data and
            lapping domains.                                     emphasize different aspects of the data
            Processing the Raw Data                              that they are presenting. Our approach
                                                                 here has been to provide a visualization
            Raw trace data by itself is not very useful.         framework (VALLEY) along with a front
            Not only is there much too much informa-             end, MIRAGE, to offer the programmer a
            tion to create meaningful displays (tex-             range of different styles. The current
            tual or graphical), but the information              system provides about ten different
            relevant to a particular problem is typi-            styles of 3D visualizations including sev-
            cally well hidden. To make the data                  eral file-based visualizations similar to
            useful, one must look at the results of one          SeeSoft, dot plots, compact trees, graphs,
            or more analyses that are run separately             time-based mappings, spiral-based linear
            or concurrently on the same trace before             views, and interval analysis.
            being combined to provide the appro-
            priate visualization data.                           The CACTI system interfaces with this
                                                                 framework by letting users select a visu-
            The current analyses we use include pro-             alization appropriate to the data they
            filing information (both single-level, à la           specified. The system analyzes the data


                                                       conduit! 11
 and the different visualization strategies              ergistic manner. Another aspect, work
 to determine which may be appropriate.                  being done jointly with David Laidlaw, in-
 The user then can select one of the appro-              volves developing new art-based visual-
 priate ones and parameterize it, speci-                 ization techniques that can convey more
 fying properties of the visualization and               information in the limited screen space
 associating data fields with visualization               available and make patterns in the data
 properties.                                             more apparent to the viewer.
 To be useful for understanding, however,                Conclusion
 even the most sophisticated visualization               This research, a continuation of the soft-
 needs to be interactive, so that the pro-               ware visualization work done at Brown
 grammer can browse over the data and                    over the past seventeen years, has been
 then focus on the most relevant aspects.                ongoing for the past six years and we ex-
 MIRAGE lets the programmer fly over                      pect it to continue for several more before
 the data, change the various parameters                 we have a truly practical system for un-
 defining the visualization, and, to a lim-               derstanding complex system behavior.
 ited extent, interact with the visualiza-               The work is funded in part by the Na-
 tion.                                                   tional Science Foundation, and we are al-
 Much of our ongoing research in this area               ways looking for additional collabora-
 involves extending these browsing tech-                 tions. Here we are interested in identi-
 niques to make our visualizations more                  fying potential users, establishing re-
 useful. Here we are looking into different              search partnerships, getting people’s
 ways of filtering the data to be visualized              feedback and experiences with software
 before it is displayed, letting the user dy-            visualization, and identifying specific be-
 namically control what is displayed or                  havioral problems that visualization
 not displayed, and correlating multiple                 might be able to address.
 visualizations of the same data in a syn-


                                                         be available for sale on diskette as down-
ALUMNI EMAIL                                             loads or as Rocket Editions. For more
                                                         information,     visit   http://www.awe-
                                                         struck.net or my web site at http://
         MARY TAFFS ’75                                  www.spiritone.com/~mtaffs/index.html.
          I’d like to announce the publica-
                                                         I fell in love with the world of books
          tion of my first novel, Martha’s
                                                         before I could read, and writing is what
          Madness, by Awe-Struck E-
                                                         I’ve always wanted to do. My computer
          Books this September. Mar-
                                                         career started out at the American Math-
          tha’s Madness is a love story at
                                                         ematical Society in downtown Provi-
          heart, but also the story of a
                                                         dence. Later, I worked at SofTech in
          woman learning to value and
                                                         Newport for several years doing Ada com-
          trust herself. It takes place on
                                                         piler development with my husband Dave
          the Oregon Coast and in Rhode
                                                         (’75). We moved to the Portland, Oregon
          Island. The hero and heroine
                                                         area in 1986, and I’ve worked most of the
          met at Brown as undergradu-
                                                         time since then at Mentor Graphics on
          ates, and the heroine was a com-
                                                         DOC, technical publishing software
          puter science major. I was very
                                                         which was discontinued as a product in
 pleased with the following recent review
                                                         1991, but is still supported for the sole
 which gave the book four stars and classi-
                                                         use of Boeing.
 fied it a “Must Read.” Said reviewer
 Angie Evans of Reviews@SimeGen, “I                      I decided in about 1993 that if I was ever
 enjoyed this delightful book. It was well               going to write, I’d better get busy NOW.
 written and the characters were believ-                 My favorite books at the time were mys-
 able. Martha’s Madness is a good read.”                 teries and thrillers, so I decided to write
 A second novel, Celtic Knot, featuring a                an amateur sleuth series. I found out,
 Brown CS grad as heroine will be pub-                   though, that while I love to read them,
 lished in a few months. Both books will                 I’m not good at coming up with all those


                                                conduit! 12
                    plot twists and red herrings, and the rela-              but it still feels familiar. Not that there
                    tionships between the characters were                    weren’t any surprises: I found a coffee-
                    what I most loved writing about. That led                cart in the lobby of the CIT, which would
                    me to move more into the women’s fiction/                 have been unimaginable five years ago.
                    romance area, and that’s where I truly                   Ah, progress!
                    belong. mtaffs@SpiritOne.com.
                                                                             I enjoy Seattle immensely, and see sev-
                                                                             eral former Brown CS folks on a regular
                    MAARTEN VAN DANTZICH ’93                                 basis, including Dan Robbins ’91 (who’s
                    Since leaving Brown I’ve been at Micro-                  in the office next to mine and one of my
                    soft Research, exploring the use of 3D                   close coworkers), Russell Belfer ’91, and
                    graphics in user interfaces for desktop                  Matt Ayers ’95/MSc ’98. And of course
                    applications. I enjoyed visiting Brown in                there are lots of Brown alums at Micro-
                    September and presenting an overview of                  soft: enough to fill a whole article with
                    our recent work; it’s fun to be the visiting             the variety of jobs we have. Maybe in a
                    presenter in Lubrano now, and great to                   future conduit!?
                    see how healthy the department is. In five
                    years many of the people have changed,



                                                                            Andy remembers John as exemplifying
          JOHN GANNON                                                       the best of Brown’s “get-involved-up-to-
                                                                            your-eyeballs” undergraduates: he was a
    (BS ’70, ScM ’72) 1948-1999                                             UTA, a research assistant, paymaster (he
                                                                            kept the books for the graphics group),
                   John Gannon, one of Andy van Dam’s
                                                                            helped write proposals and critiqued ev-
                   first and favorite students and one of
                                                                            eryone’s work. John had a very idiosyn-
                   CS’s most distinguished graduates, died                  cratic laugh (a wicked cackle) easily
                   in his sleep of cardiac arrest at his home               evoked; he never took himself or others
                   in Silver Spring, MD, on June 12. He                     too seriously, but he had a great serious-
                   was 51.                                                  ness of purpose. He was always reliable
                                            John had a con-                 and just got stuff done, no excuses—he
                                            genital heart de-               had a “My dog ate it” stamp made to use
                                            fect, but surgery               on the programs of students reduced to
                                            as a child had                  creatively lame excuses for late or incom-
                                            made it possible                plete submission.
                                            for him to live an
                                                                            At the time, Andy and Peter Wegner were
                                            active and pro-
                                                                            the two people teaching CS in the Divi-
                                            ductive life. He
                                                                            sion of Applied Math (John Savage was
                                            was Chair of the
                                                                            teaching in the Division of Engineering)
                                            Department       of
                                                                            and Andy was trying to cover too many
                                            Computer       Sci-
                                                                            aspects at once via AM101 and 102. He
                                            ence at the Uni-
                                                                            included a little bit of parsing and com-
                                            versity of Mary-                piler theory, a subject then still in its in-
                                            land, where he                  fancy. One of the things John Gannon and
                                            earned a reputa-                Andy did was to try and make sense of
                                            tion as a leading               Frank DeRemer’s brand-new PhD thesis
                                            researcher      in              on parsing to see what of it they could
                                            software     engi-              teach to undergrads. John read the thesis
                                            neering       and,              and then would try to explain the basics
                                            most       impor-               to Andy; Andy would find holes in the ex-
                                            tantly to him, as               planation, and back John would go to try
                                            a demanding but                 and work it out. After many such cycles
John and his wife Nancy participating       caring    teacher               they doped it out and managed to teach
        in Andyfest last May                whose door was                  hot-off-the-press LR(k) parsing to mere
                                            always open.                    sophomores and juniors.


                                                                   conduit! 13
                   Doing this digging had a life-long impact               cation at the University of Maryland. In-
                   on John. He went to the University of                   formation about this is available via
                   Toronto for his PhD, as did many of his                 http://www.cs.umd.edu gannon_memor-
                   generation from Brown, including Ed                     ial.html. Condolences may be sent to
                   Lazowska, Frank Tompa, Chris Braun,                     John’s family at this email address:
                   Larry Weissman, David Elliott, John                     gannon-memorial@cs.umd.edu. John, who
                   Guttag, and John Zahorjan—a pipeline                    was raised in Rhode Island, is survived by
                   from Brown to Toronto that continued for                his brother, Rick, of Foster, and by his
                   many years. In his professional career,                 wife, Nancy Garrison ’70, a Yale Law
                   John became a real authority on software                graduate who works at the U.S. Depart-
                   engineering and brought rigor to an area                ment of Justice. Nancy can be reached at
                   sometimes lacking it. He also became a                  10108 Day Avenue, Silver Spring, MD
                   superb teacher and mentor.                              20910.
                   The John D. Gannon Scholarship Fund                     John’s many friends and colleagues at
                   has been established to commemorate                     Brown mourn deeply his untimely
                   John’s commitment to students and edu-                  passing.



                                                                           sought contains many fewer words, and
      THE 23rd IPP SYMPOSIUM                                               thus is less likely to contain exactly the
                                                                           words in the query. Woods looked at how
                                                                           various natural-language technologies
                       The Computer Science Department                     can help in this problem. The techniques
                       held its 23rd Industrial Partners                   he considered ranged from the relatively
                       Program Symposium (less formally                    basic to those closer to the research edge.
                       known as IPP day) on April 29, 1999,                At the basic level I was surprised to learn
                       on ‘‘Web-based Natural-Language                     that ‘‘stemming’’ (reducing a word to its
                       Technology: Search, Translation and                 basic form—e.g., the word ‘‘reducing’’ be-
                       Analysis.’’ As usual, John Savage,                  comes ‘‘reduce’’) has a major positive im-
                       who runs the program, warmed up                     pact in retrieval rates. Among the more
                       the audience by having everyone in-                 complicated techniques described was
                       troduce themselves and in the pro-                  ‘‘subsumption’’—finding concepts that are
                       cess demonstrated that he knew the                  either more general or more specific than
                       names of pretty much everyone                       a given concept. The idea here is that if
Host Eugene Charniak there (which, from my perspective, is                 the user asks about, say, ‘‘computer’’
                  an amazing feat). He then turned things                  prices, we might also want to give infor-
                  over to me, and I promptly introduced the                mation about ‘‘workstation’’ prices, and
                  first speaker (whose name, fortunately, I                 vice versa.
                  did know: Bill Woods).
                                                                           The second talk of the day was by Roy
                   Bill Woods is from Sun Microsystems                     Byrd of IBM Watson, on ‘‘Text Mining for
                   Labs and has been doing work on using                   Knowledge Management.’’ Roy pointed
                   natural-language technology to improve                  out that one of the most profitable areas
                   information retrieval. His talk was enti-               of IBM today is its consulting business.
                   tled ‘‘Finding What You Really Want:                    This business generates a lot of informa-
                   Natural Language Technology in Preci-                   tion about what IBM customers are doing
                   sion Content Retrieval.’’ The standard                  and the problems they are having doing
                   technology in this area is to start with the            it. This information is both a problem and
                   set of words in the user’s query and re-                an opportunity. Essentially Roy and his
                   turn documents that contain those same                  group at IBM are working on an on-line
                   words. However, usually users do not                    document creation and analysis package
                   want an entire document, but rather just                to be used by the IBM consulting division
                   a passage that answers their question.                  to find documents that relate to their que-
                   This is the “Precision Content Retrieval”               ries, route documents automatically to in-
                   of the title, and it is much harder than                dividuals who could use the information,
                   document retrieval since the passage now                and use the entire document collection to


                                                                  conduit! 14
          gather global information about trends in             lates to language issues. However, his
          the area. One of the technologies Roy                 talk today was not so much on language
          stressed was the use of finite-state au-               as on questions about computer design in
          tomata. Natural language is funny in                  the macro sense—if one is to walk around
          that few of the tasks involved can be com-            with a computer, how big should it be,
          pletely solved within the domain of finite-            how should one hold it or otherwise carry
          state machines, but a huge number of                  it, etc. If the audience ever noticed that
          them can be approximated using these                  the talk did not relate all that closely to
          techniques. The advantage of doing so is              the topic of the symposium I am sure they
                                                                quickly forgot—I know I did. This was the
                                                                only talk I have ever seen where the
                                                                speaker came equipped with fifteen to
                                                                twenty computers and proceeded to pull
                                                                them out of his backpack and discuss the
                                                                design issues involved in their creation
                                                                and why they did or did not work as prod-
                                                                ucts. One interesting point Ted made con-
                                                                cerned computers used by people gather-
                                                                ing interviewing data. It turns out that if
                                                                the computer is designed so that only the
                                                                poller can see the screen, people are much
                                                                less willing to talk than if they can see the
                                                                screen at least part of the time. More gen-
                                                                erally, Ted emphasized that one has to
                                                                think long and hard about how a com-
                                                                puter is to be used in the real world before
                                                                coming out with a product, and that most
                                                                of the machines he pulled from this back-
                                                                pack never got produced commercially be-
                                                                cause they failed this test.

                                                                The day’s penultimate talk was by Steven
                                                                DeRose, who founded Electronic Books
Symposium speakers, clockwise from Eugene:                      Technology (now Inso eBusiness Technol-
  Ted Selker, IBM; Steve DeRose, Brown; Bill                    ogies), but recently returned to Brown
         Woods, Sun; Roy Byrd, IBM
                                                                halftime in the Scholarly Technology
          that finite-state technology is very robust            Group. His talk was entitled ‘‘Links, Que-
          and very efficient. Roy gave some num-                 ries and Language Awareness on the
          bers on the speed of processing that were             Web.’’ One way to think about this talk is
          quite amazing. The talk also stressed                 that whereas the other talks were about
          that the goal here was a system that                  making computers better consumers of
          would be available in the near future but             text, this one was about making the com-
          would, at the same time, incorporate a                puter’s job easier by having the human
          very large number of language tools. The              text producer take on the burden of
          way to meet both of these desiderata is to            marking up the text to indicate some of its
          keep each tool fairly simple (which dove-             meaning. That is, we want marks that in-
          tails with the use of finite-state tech-               dicate not just how to display some text,
          nology).
                                                                but also indicate something about its con-
          The first talk after lunch was given by                tent. Steve first noted that hypertext
          Ted Selker, also of IBM, and an IBM                   markup language (HTML) goes a little
          Fellow. Ted is known, among other                     way in this direction, but not nearly as far
          things, for inventing the track-point de-             as most people would like. The bulk of the
          vice on IBM Thinkpads. Ted is also a                  talk then was on XML (eXtensible
          Brown alumnus and took my AI course as                Markup Language), which has markup
          an undergraduate. I invited him because               symbols for more content types and a
          he is well known as an interesting                    standard way to extend the mark set to
          speaker and he has done work that re-                 your domain. One important point here—


                                                       conduit! 15
I think Steve said it, but if not it was cer-        New York Times and parse every sentence
tainly implicit in his talk—is that there is         therein, even with numerous mistakes.
going to be a strong incentive to add these          Today we have several such programs (al-
markups to very many web document                    most all statistically based), and now the
producers: the web search engines at                 research effort is to drive down the
some point are going to be sensitive to              number of errors the programs make. The
XML markups, and thus if you want them
                                                     other topics in the talk were more se-
to put your web site at the top of the list
                                                     mantic in nature. Even though statistics
for a user’s query, you had better put in
                                                     are easiest to gather on surface phe-
markups that allow the search engine to
                                                     nomena like words, statistical techniques
‘‘realize’’ your site contains the informa-
tion the user has requested (e.g., the price         are beginning to be applied to the less
of a new Ferrari).                                   surfacy area of ‘‘meaning.’’ So in the work
                                                     on pronoun reference we took some text,
The last talk of the day was mine, on ‘‘The          marked it up to indicate the referents of
Statistical Revolution in Natural-Lan-               all the pronouns, and then gathered sta-
guage Processing.’’ My goal here was not
                                                     tistics on such things as the probability of
to illustrate how this technology can be
                                                     the pronoun’s antecedent being, say, N
used (I figured that the other speakers
                                                     sentences back, for N = 0, 1, 2, etc. One in-
would be able to do this far better than I),
but simply to show how the use of statis-            teresting subproject here relates to the
tics has revolutionized computational lin-           fact that pronouns in English have a
guistics, and how it has led to remarkably           gender that must match the gender of the
robust and accurate programs that attack             antecedent. Thus a goal was to automati-
a wide variety of natural-language topics.           cally collect information about the typical
In particular, I talked about four topics:           gender of objects described by words like
parsing, word-sense disambiguation, pro-             ‘‘piano’’ (neuter), ‘‘president’’ (mostly
noun reference, and lexical semantics. I             male), etc. This project relates to the lex-
find the work in parsing (assigning a sen-            ical-semantics portion of the talk, as the
tence structure to a string of words) par-           goal of statistical lexical-semantics is to
ticularly exciting. As recently as six or            learn semantic information about words
seven years ago, there was no parser that            by observing how they occur in text.
could take, say, the front page of today’s




    A recent WiCS (Women in Computer Science) meeting at which Philip Klein was
  invited to speak. WiCS acts as a resource for women in the department and offers a
       mentoring program and a ‘safe space’ for women to discuss related issues


                                            conduit! 16
                         The astaff team is a blur of activity during a stuffing blitz to mail out our
                          graduate recruitment posters. l to r: Fran Palazzo, Genie deGouveia,
                                      Lori Agresti and the back of Dawn Nicholaus



                                                                      for philosophy. During their time at the
           NEW CS FACULTY                                             University of Bonn, they also spent a
                                                                      year in Paris, one of the great centers for
                                                                      contemporary philosophy, which helped
                                The Department is very                coalesce their future intellectual direc-
                                pleased to be welcoming               tions. Despite having taken myriad phi-
                                three new faculty mem-                losophy courses, Thomas decided upon a
                                bers this year—William                career in CS. Vera received her PhD in
                                Chan, PhD University of               philosophy. Their respective degrees were
                                Washington, 1999; Amy                 awarded within days of each other and
                                Greenwald, PhD Cou-                   they came to Boston, Thomas as a
                                rant Institute, New York              postdoc at MIT and Vera continuing her
                                University, 1999; and                 research at BU. Their son Emmanuel was
                                Thomas Hofmann, PhD                   five at the time and had to learn English
                                University     of   Bonn,             fast. He now speaks English very well—
                                1997. Professors Chan                 he with an American accent, his parents
                                and Greenwald won’t ar-               with an Oxford one. The challenge now is
                                rive until the new year               to make sure he retains his German.
                                and will start teaching in
                                Semester II, so news                  From Boston, they headed for the west
                                about them will be forth-             coast and a year at Berkeley—Vera in the
                                coming in the spring                  Rhetoric department, Thomas in CS.
                                issue of conduit! Pro-                They enjoyed Berkeley’s colorful ethnic
Vera, Emmanuel and Thomas       fessor Hofmann is here                diversity, culture and climate. From
 Hofmann at the CS picnic                                             thence they returned to the east coast
                                and teaching an ad-
                vanced topics course this semester on in-             and Brown. Vera now has a lectureship
                formation retrieval and data mining.                  in philosophy at URI and Emmanuel is
                                                                      now in second grade in Barrington. He
                  Thomas and his wife Vera grew up in a               especially enjoys the school bus—he’s the
                  town near Cologne. They met in high                 first student picked up, so he’s aboard the
                  school and discovered a mutual passion              longest.



                                                             conduit! 17
Thomas’ research goals can best be sum-                mation theory, applied mathematics, sta-
marized by the motto “artificial intelli-               tistical physics, electrical engineering,
gence by machine learning.” He is                      operations research, and computational
interested in how computers can take ad-               linguistics. One of his main motivations
vantage of large amounts of data in order              for coming to Brown is its strong culture
to achieve better performance for a va-                of interdisciplinary research and its op-
riety of AI-related tasks. His previous                portunities for collaboration across de-
work focused on problems in pattern rec-               partmental boundaries.
ognition, machine vision, information re-
trieval, data mining, and natural lang-                Philosophy is still a passion that often
uage processing. Here, the term “data”                 finds Thomas reading and arguing into
spans a large spectrum from raw mea-                   the night. Vera is, of course, a worthy op-
surements and sensory inputs at one end                ponent in debate and there’s always
to discrete and symbolic data at the other             plenty to discuss. Thomas enjoys writing
extreme. Some of the key questions that                essays combining philosophy and com-
fuel Thomas’ re-
search are: What are
the principles of in-
                          “One of his main motivations for
ductive     inference,  coming to Brown is its strong culture
i.e., how can we infer
general laws from a
                        of interdisciplinary research and its
set of examples?           opportunities for collaboration
What are the mecha-
nisms that would en-     across departmental boundaries”
able machines to
understand and interpret images? How                    puter science, analyzing CS’s impact on
can computers process and understand                    society and the ensuing societal changes.
natural language? What methods can be                   Chess is another major interest. He plays
devised to automatically detect structure               with friends in Germany via the Internet,
and regularity in large data sets? How                  and thanks to an enrichment program
can data analysis and data visualization                run by the California school system, Em-
be combined to make human-computer                      manuel is also able to play. Vera is an ac-
interaction more efficient? How can com-                 complished pianist and church organist.
puters support humans in solving deci-                  She misses being able to play the organ
sion problems in complex and uncertain                  in her church back home to which she
environments? While he is greatly inter-                had a key and could practice pretty much
                                                        at will. Fortunately for all, their piano re-
ested in the mathematics and theoretical
                                                        cently arrived at their new home in one
foundations of machine learning, he has
                                                        piece.
also a profound interest in modeling and
solving specific problems. It is his strong              As a former Green Party activist, Thomas
belief that the development of new                      avoids driving his car and opts instead
models and methods can greatly profit                    for traveling by bike or public transport
from a confrontation with real-world                    whenever possible. He uses the East Bay
problems, while on the other hand “there                Bike Path to come to work each morning
is nothing more practical than a good                   (as does David Laidlaw). Since he’s now
theory” (V. Vapnik). The nature of these                in Rhode Island for the long haul, he will
                                                        likely become involved in politics again.
problems often requires a cross-disci-
plinary approach, involving various as-
pects of computer science from theory to
computer graphics to systems design as
well as disciplines like statistics, infor-




                                              conduit! 18
                                                     fac.activities@cs.brown.edu
                    Tom Doeppner. Tom enjoyed visiting                     was keynote speaker at a workshop in
                    Swami Manohar, PhD ’89, and his family                 theoretical CS at the IBM T.J. Watson
                    last January in Bangalore, India. He was               Research Center.
                    there to teach a course on distributed
                    computing for CS Industrial Partner                                     www
                    Compaq.

                                      www                                  Steven Reiss. Along with David Laid-
                                                                           law, Steven has been awarded an NSF
                                                                           grant for studying software visualiza-
                    David Laidlaw. David’s new course on                   tion, as described elsewhere in this issue.
                    interdisciplinary scientific visualization              He is also teaching CS233, where this
                    is centered around writing mock grant                  semester the class is attempting to build
                    proposals, reviewing them by emulating                 a modern Java programming environ-
                    the NSF review process and recom-                      ment for the Suns. To keep him busy
                    mending proposals for ‘funding.’ Prac-                 beyond these activities, he continues to
                    ticing what he preaches, David is co-PI on             work on providing tools for using design
                    a new KDI grant from NSF, “3D Free-                    patterns throughout the development
                    Form Models for Geometric Recovery and                 process, designing user interfaces for
                    Applications to Archaeology,” with David               mobile computing, developing a front end
                    Cooper from Engineering as PI. He and                  for searching the Internet with a local
                    Steve Reiss were just awarded an NSF                   startup company, Simpli.com, and har-
                    grant for visualizing program structure                vesting his vegetable garden.
                    and execution to help in understanding
                    how large programs work (or don’t
                    work...).
                                                                                             www

                                     www                                   Roberto Tamassia. Roberto gave a
                                                                           keynote lecture at the VIII Encuentros de
                                                                           Geometría Computacional in Castellón,
                    Franco Preparata. Franco was re-                       Spain. His book, Data Structures and Al-
                   cently appointed chair of the Gödel Prize               gorithms in Java (coauthored with
                   Committee. It is the most prestigious                   Michael Goodrich), has reached its 6th
                   award in theoretical CS for an out-                     printing.
                                       standing      paper(s)
                                       published in the pre-
                                       vious six years. In                                   www
                                       July he served on an
                                       international       re-
                                       view committee for
                                                                           Eli Upfal. In May Eli and PhD candi-
                                                                           date Gopal Pandurangan went to the
                                       the Department of
                                                                           31st ACM Symposium on Theory of Com-
                                       CS of the Univer-
                                                                           puting in Atlanta, where they presented
                                       sity of Pisa, Italy. In
                                                                           a joint paper. In July, Milos Hauskrecht
                                       August Franco pre-
                                                                           (Eli’s postdoc) presented a joint paper
                                       sented several lec-
                                                                           with Gopal and Eli at the 16th Interna-
                                       tures at a summer
                                                                           tional Joint Conference on Artificial In-
                                       forum on supercom-
                                                                           telligence (IJCAI) in Stockholm. He was
                                       puting organized at
                                                                           appointed editor-in-chief of the Journal
                                       El Escorial, Spain,
                                                                           of Discrete Algorithms.
                                       by the Universidad
                                       Complutense,         in
                                       Madrid. Recently he
The Manohar triplets sporting t-shirts                                                        www
      from Tom Doeppner


                                                                 conduit! 19
              Andy van Dam. Andy co-chaired the                   nical advisory board for Microsoft Re-
              first joint European Commission/Na-                  search, Andy went to Beijing in June,
              tional Science Foundation (EC/NSF) ad-              where Microsoft has recently opened a re-
              vanced research workshop, entitled                  search lab. Andy was part of a contingent
              “Human-Centered Computing, Online                   including Raj Reddy (CMU), Ed
              Communities and Virtual Environments,”              Lazowska (Chairman of CS at the Uni-
              held in France. As a member of the tech-            versity of Washington), and others that
                                                                  provided a full day of lectures on the fu-
                                                                  ture of computing to some 1500 invited
                                                                  academics and students—the first such
                                                                  gathering in China. Andy’s talk was on
                                                                  Post-WIMP 3D User-Computer Interac-
                                                                  tion.

                                                                                   www

                                                                  Stan Zdonik. In September Stan com-
                                                                  pleted his duties as North and South
                                                                  American Program Chair for the 25th In-
                                                                  ternational Conference on Very Large
                                                                  Databases (VLDB) held in Edinburgh,
                                                                  Scotland.


                                                                  GRAPH papers were “(Gimme some)
  SIGGRAPH 99 SHOWCASES                                           skin: A constructive approach to free-form
                                                                  modeling” and “Art-based rendering of
       BROWN WORK                                                 fur, grass, and trees.” The first of these
                                                                  (by Lee, Jon Cohen ’00, J.D. Northrup ’00,
              In LA this August about 50,000 people
                                                                  and me) described how to build large and
              gathered at SIGGRAPH 99 to see what’s
              new in computer graphics. There were tu-            complex free-form shapes (like the torso
              torials, a spectacular “show floor” with all         shown in Figure 2) by a technique based
              the latest and greatest new products,               on constructive drawing in which one de-
              panels, a “digital bayou,” a film and video          fines a form by first placing large shapes
              show, panels on controversial topics,
              and technical paper sessions. At least
              25 people from Brown attended and
              participated in several venues. I’m not
              going to give the details of all of
              them—but I will describe the technical
John Hughes   papers we presented, which is the area
              I’m most familiar with.
              PhD student Lee Markosian was lead
              author on one paper and coauthor on
              another, both originating from his
              work on art-based modeling and ren-
              dering, to become his dissertation this
              spring. The overall goal of this work is
              to incorporate into computer-generated
              images the kinds of skills that artists
              have—representing complexity with
                                                        This picture, a computer-generated rendering
              just a few strokes of a brush or pen,       of a Dr. Seuss-like scene, was chosen for the
              creating 3D shapes through “construc- cover of the SIGGRAPH Proceedings. This con-
              tive drawing,” and selectively using de- tinues a trend: Brown students, faculty or gradu-
              tail to direct attention to what’s ates have been authors of papers featured on
              important in an image. Lee’s SIG-          SIGGRAPH covers in four of the last five years


                                                         conduit! 20
                like cylinders or ellipsoids, cubes, and             behave nicely elsewhere. “Nicely” means
                other “primitive” objects and then con-              that it avoids self-intersection and tries
                structing a “skin” around them. The skin,            to avoid wrinkles or creases, and degen-
                in this case, is an adaptively tesselated            erate triangulations with long, thin, tri-
                mesh, positioned to have a user-chosen               angles. The speed with which one can
                offset from the underlying primitives in             create complex free-form models, espe-
                places where a primitive is nearby, and to           cially when one has Bob Zeleznik’s
                                                                     SKETCH system to help with placing the
                                                                     primitives, is remarkable.

                                                                     The other paper got to be informally
                                                                     known in the graphics group as “fuzzy fur
                                                                     feet,” because the first successful proof-of-
                                                                     concept picture was based on a drawing
                                                                     from Dr. Seuss’s The Foot Book. The
                                                                     paper was written by Michael Kowalski
                                                                     (BA ’98, Scm ’99), Lee Markosian, J.D.
                                                                     Northrup ’00, Lubo Bourdev (ScB, Scm
                                                                     ’98), Ronen Barzel (ScB ’83), Loring
                                                                     Holden, and me. In an early draft of his
                                                                     presentation, he spent the first few min-
                                                                     utes explaining why “non-photorealistic
                                                                     rendering” was useful. Fortunately, some-
                                                                     one who listened to it said “Michael: if
                                                                     they’re there to listen to you talk in the
Figure 2: Jon Cohen, who had never used a commercial                 last session of the last day of the confer-
 modeling system and had no background in anatomy,                   ence, I think you can assume that they
was assigned to produce a torso with our system. His first            believe what you did had a purpose!” I’m
one looked terrible (it had no collarbones, for example),            pleased to say that a lot of people were
and I told him to try again. And then again. And then he             there—about 1500 of them—and that
  got an anatomy text from the library, and tried again.             Michael’s presentation was a real suc-
  And again. Less than 24 hours after he started (with a             cess. Our work on this topic isn’t done
short sleep break), he produced the model shown here                 yet: Michael, now working at ATR in
                                                                     Japan, is collaborating with us on exten-
                                                                     sions to this work and another project on
                                                                     art-based graphics. Look for us at SIG-
                                                                     GRAPH next summer!




                                                    Figure 3: A car-
                                                    toon-style foot
                                                      with creases
                                                  between the toes.
                                                  This sort of model
                                                   cannot easily be
                                                  made with implicit
                                                   surfaces or with
                                                     conventional
Figure 4: With the same technology used to
                                                  modeling software
  create the cover image, we also make
Durer-like images, with dark and light strokes
 used to convey highlights and rich texture

                                                            conduit! 21
                                                                   Over the summer, President Gordon Gee
                                                                   visited General and Mrs. Kanellakis at their
                                                                  home in Athens to thank them for establish-
                                                                  ing graduate fellowships in memory of their
                                                                  son Paris and his family and for their support
                                                                   of CS faculty. There was an exchange of
                                                                    gifts—a gloriously illustrated book about
                                                                  Greece from the Kanellakises and for them,
                                                                     a tie and a scarf specially designed as
                                                                             Brown presentation gifts



                                                                          eminent multiethnic society, can draw
      CHARNIAK UNPLUGGED                                                  the best and brightest from around the
                                                                          world. No doubt this is in part because of
                                                                          the salaries we pay, but I think few
                      Recently a newspaper article ap-
                                                                          would disagree that an equally impor-
                      peared in my e-mail on the difficulty
                                                                          tant reason is the extent to which our
                      universities are having in recruiting
                                                                          culture is open to people of all races and
                      and retaining computer science fac-
                                                                          ethnicities. (Which is not to say we are
                      ulty. This got me to thinking about a
                                                                          perfect, just much better than most ev-
                      number of such articles I have seen
                      over the last year.                                 eryone else.) Yet at the same time, I did
                                                                          not like the idea of increasing the visa
                      The first of these was about a year                  quotas. As a professor of computer sci-
                      ago, when high-tech companies were                  ence, I am disappointed that so few
                      lobbying Congress to increase the                   American students go into the sciences,
                      number of special visas given to for-               and while I do not have any data on this
                      eign workers in areas in which there                subject, my impression is that despite
                      was a shortage in the US. These com-                the rising salaries high-tech workers can
Eugene Charniak   panies said that they were not able to hire             command, it is still the case that doctors,
                  high-tech workers, particularly in com-                 lawyers, and business executives out-
                  puter science and related areas, and thus               earn computer scientists. If this is true,
                  it was important for the US to allow more               then the recalcitrant students are mak-
                  such people to come into the US from                    ing reasonable decisions. Why study
                  other countries. My thoughts on this were               something hard, like science, when you
                  somewhat ambivalent. On the one hand, I                 can major in something that is much less
                  am proud that the US, as the world’s pre-               work and still earn more over your life-


                                                                conduit! 22
                                                                           expressed no opinion on what should be
                   time?1 Since we live under the sway of
                                                                           done about the problem, but a rather
                   the laws of supply and demand, not al-
                                                                           large number of people were quoted who
                   lowing more foreign workers in should
                                                                           thought that the best solution would be
                   cause salaries to rise even further and, I
                                                                           to bring back the draft. They pointed out
                   hope, attract more US students into the
                                                                           several benefits beyond the obvious one,
                   sciences.
                                                                           that the military would not have to
                   I was still mulling this over a few months              worry about where their next recruit was
                   later when I saw a second article, this                 coming from. One was that the young
                   time about problems the military was                    would have the experience of helping
                   having in attracting enough new recruits.               their country in an important way. An-
                   As this was a “news” article, the author                other that I remember particularly was
                                                                           that this would allow the civilian popula-
                   1. Our editor, Trina Avery, takes exception             tion to better understand military cul-
                   to the implication here that science is                 ture, and would at the same time help
                   harder than non-science. Your humble au-                bring military culture into line with the
                   thor stands by his prejudices.                          mores of the average civilian.



     I’d like to thank Shuang Ji for his very generous donation of $15,000
     as an unrestricted gift to the department. Shuang completed his
     Master’s degree working with Steve Reiss in ’94. His gift comes at a
     time when the department is experiencing considerable growing pains
     as enrollments soar, the number of faculty increases, and the need for
     equipment and space outstrips the administration’s largesse. It’s espe-
     cially gratifying when a former student steps forward to help out the
     department, and we are particularly grateful to Shuang for his gift and
     the warm sentiments with which it was given.
                                                         Tom Dean, Chairman



...and one last glimpse of Andyfest




         With his father, Levie van Dam                      Most of the participants pose in front of the
                                                                          Sciences Library


                                                                 conduit! 23
                            My first reaction was that the author and               the shortage of computer science profes-
                            the quotees were on to something impor-                sors. As I saw it from my previous read-
                            tant here, but that they were not think-               ing, there were three ways to attack this
                            ing big enough. Here, I thought, was the               problem. Increasing the quotas for for-
                            solution to the computer programmer                    eign professors of computer science
                            shortage as well. We should draft people               would probably not do any good. From
                            into Microsoft, Sun, Compact, SGI, Intel,              my experience here at Brown, there does
                            etc. Not only would this solve the pro-                not seem to be any effective quota here.
                            grammer shortage, it would give the                    The typical person we wish to hire has
                            young a good feeling for having helped                 unique credentials, a case can be easily
                            the country surmount a national prob-                  made that there is no person in the US
                            lem, it would help the civilian world un-              with the same credentials, and, ... well,
                            derstand the nerd culture, and finally, it              you get the general drift. This left the
                            might even bring the culture inside of                 other two options: raise the salaries of
                            (say) Microsoft or Sun in line with that of            professors of computer science, or use the
                            the rest of the country.                               draft. On this issue, as you might im-
                                                                                   agine, I have very firm opinions.
                            It was with all this baggage rumbling
                            around in my head (please excuse the
                            mixed metaphor, but it best describes my
                            state of mind) that I read the article on




                               conduit!
                                 A publication of
                         The Computer Science Department
                                 Brown University
                                       k
                                  Inquiries to: conduit!
                           Department of Computer Science
                              Box 1910, Brown University
                                 Providence, RI 02912
                                   FAX: 401-863-7657
                                PHONE: 401-863-7610
                              EMAIL: sjh@cs.brown.edu          Suzi Howe                Katrina Avery      Jeff Coady
                           WWW: http://www.cs.brown.edu/                                                Technical Support
                                                             Editor-in-Chief                Editor
                                publications/conduit/



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

              conduit!




Printed on recyled paper                                                                           Address changes welcomed

                                                                          conduit! 24

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:12
posted:11/2/2011
language:English
pages:24