conduit!

Volume 12, Number 1 Department of Spring, 2003

most ten to fifteen years, the normal length of time for research to reach the marketplace. COMPUTATIONAL The resolution achievable through photoli- NANOELECTRONICS thography is determined by the wavelength of electromagnetic radiation used, and we are al- ready operating at high frequencies and small The Integrated wavelengths. Circuit The Future Lies with Moore’s Law postulates that the Nanotechnology number of transistors in an inte- grated circuit (a chip) will double ev- Fortunately, nanotechnologies appear to have ery few years. Remarkably, this the potential to supplement photolithography forecast of exponential growth has and allow chips of increasing density. So what been accurate for more than 45 is nanotechnology? Nanotechnology is the years. The result has been a rapid manufacture of technologies whose smallest decline in the cost of computers and dimension is on the order of a few nanometers. a rapid increase in their speeds and Chemists have been able to grow carbon nano- storage capacities. This technologi- tubes (CNTs), tubes made of carbon atoms, and cal revolution has led to productivi- semiconducting nano-wires (NWs), crystalline ty gains of historic proportions. wires made of a semiconductor material, both of which make good wires. Diodes and transistors Moore’s Law is a consequence of can also be formed from CNTs and NWs. Thus, continuing advances in photolithog- the building blocks for computers can be realized raphy, the process of inscribing John Savage from nanomaterials. Nanotechnology chip manu- transistors and wires on the sur- face of crystalline semiconductor materials. Photolithography is a Today photolithography is multi-step process in which light passes through masks whose open- complex and precise. Features ings specify the location of wires as small as 90 nanometers and devices and falls onto photo- -7 sensitive coatings on a semicon- (10 meters) can now be ductor wafer. Solvents then remove the areas exposed to light imprinted on chips so that they can be used for wires and devices. facture is expected to use self-assembly, a pro- cess in which very large numbers of objects with Today photolithography is complex and pre- nanometer dimensions arrange themselves into -7 cise. Features as small as 90 nanometers (10 stable configurations. Chemists are adept at meters) can now be imprinted on chips. This manufacturing molecular structures using sophistication is costly; the ultraclean facto- self-assembly. However, since this process is ries required to manufacture such high-densi- primarily stochastic in nature, while nano- ty chips cost several billion dollars to technology chips are very likely to exhibit a construct. Unfortunately, it is predicted that high degree of regularity, the exact location of the exponential growth in chip capacity individual objects, such as CNTs and NWs, through photolithography will end within at probably cannot be controlled.

Brown University, Box 1910, Providence, RI 02912, USA Crossbars, An Important Lincoln of SRI and I have proposed a way to Architectural Element construct an address decoder solving this problem that uses the fact that p-type re- So we ask, what types of computers are like- gions can be doped into NWs along their ly to be built using nanotechnologies? The length. Each region is conducting unless a current thinking is that crossbars, a two-di- sufficiently large positive electric field (de- mensional array (nanoarray) formed by the noted by 1) is applied above. Thus, a wire intersection of orthogonal sets of parallel placed above or below a region at right an- and uniformly spaced NWs organized into gles forms a field-effect transistor (FET) rows and columns, will be an important that acts like a switch and is turned off when technology (see Figure 1a). a sufficiently large positive voltage is ap- plied on an orthogonal wire. We use this re- Researchers have shown that nanowires can sult to design a decoder. be aligned in parallel with nanometer spac- ings using directed self-assembly and that We assign to each NW a binary code, called an h-hot address, con- taining h 1s and (b – h) c1 c2 c3 c4 c5 c6 nw1 nw3 nw5 0s. We assign a 1 to an nw nw nw2 4 6 undoped region and a 0 r1 to a doped region. We propose that coded aw 1 NWs be laid down at r2 aw right angles and on top 2 of a set of microwires r 3 aw3 that overlap the doped and undoped regions, r 4 aw4 as suggested in Figure r 1b. If all possible codes 5 appear among the NWs, applying a high r6 voltage to h of b mi- crowires will result in only one NW that is (a) (b) conducting. This allows each NW to be Figure 1: (a) A nanoarray with dots showing locations of 1s; 0s are in the other locations. addressed. (b) 2-hot addressing of six nanowires {nw1,...,nw6} with four microlevel address wires Given the stochastic na- {aw1,...,aw4}. In practice the microwires would be much larger than the nanowires ture of directed self-as- sembly, it cannot be two such layers of nanowires can be assem- ensured that every differently coded NW bled at right angles. They suspend NWs, will appear. Similarly, NWs are unlikely to which are so small as to be invisible by a align perfectly with an orthogonal set of mi- light microscope, in a solution and float crowires. We have shown how to cope with them through a trough on a chip. The NWs these uncertainties to design a decoder that orient themselves as would logs in a stream can be used in a nanoarray. Ours is the sec- and come to rest aligned in the direction of ond decoder for NWs to be proposed; the the flow. Materials have also been developed first, based on the random deposition of gold that permit nanoarrays to store binary data nanoparticles on the intersections of the at a crosspoint, the intersection of a pair of NWs and microwires, has its own set of man- orthogonal nanowires. Thus, nanoarrays for ufacturing challenges. Our decoder provides data storage are a realistic possibility. an important technique for bridging the gap Addressing Nanoarrays between microscale and nanoscale features and for bootstrapping the programming and Access to nanoscale devices will necessarily customization of nanoscale systems. be from the micro level. If nanowires must be attached individually to microwires, no Storing Data in Nanoarrays advantage will accrue to nanotechnology. Nanoarrays can be used for data storage. Re- Thus, to use nanoarrays efficiently some searchers have developed molecular films method must be found to address a large that, when placed between the row and col- number of NWs using a small number of mi- umn NWs of a nanoarray, become conduct- crowires. André DeHon of Caltech, Patrick ing (nonconducting) at the intersection of

conduit! 2 rows and columns when a large positive (neg- obtained in polynomial time. If only stores ative) voltage is applied between these rows were available to program nanoarrays, we and columns. Thus, if a large positive (nega- might conclude from these results that the po- tive) voltage is applied across a set R of rows tential of nanoarrays is indeed limited. Fortu- and a set C of columns, the film in the regions nately, for some prototype problems nano- around the crosspoints formed by these rows arrays can be programmed much more effi- and columns becomes conducting, represent- ciently with stores and restores than with just ed by 1s (nonconducting, represented by 0s). stores alone. For example, an nn× } nanoarray This property means that nanoarrays should with 1s on the diagonal requires n stores but be able to store large amounts of data. We say only 2log2n stores and restores. Although this that a store (restore) operation is done when 1s suggests that the problem of minimizing the (0s) are written. Nanoarrays can also be used number of operations might not be NP-hard, to compute, since one row or column can be we have developed a complex proof that the seen as computing the “wired OR” of its input. problem remains NP-hard. Since inverters can be constructed using NWs, every Boolean function can be realized The above results apply when each row and by these nanoarrays. column NW can be individually addressed. Since nanoarrays accessed from the micro-lev- Lee-Ad Gottlieb, Arkady Yerukhimovich and I el will probably use a form of h-hot address- have considered the following questions: a) What are the most efficient ways of entering ing, we have also examined the same data into large nanoarrays? b) How difficult is questions when h-hot addressing is used. it to find a minimal or near-minimal number While individual nanowires can be controlled via decoders, decoders also constrain the set of of stores and restores to enter the data? wires that can be addressed simultaneously. We have shown that the answer to the second As a consequence, more time will be needed to question when the normal data entry method enter data into a nanoarray in this case. Be- is used, i.e. when only stores are performed cause of the diffculty of programming nanoar- (restores are not used), is that it is NP-hard. rays with h-hot addressing, we have explored In addition, we have shown that it remains the programming of individual rows or col- NP-hard to obtain good approximations to the umns. We have shown that it is NP-hard to minimal number of operations to program a find a minimal or near-minimal number of nanoarray. However, when a fixed upper steps to program a single row or column of an bound is placed on the number of 1s in rows or array. However, we have also shown that this columns, although the program remains NP- problem is log-approximable when h and b are hard, reasonably good approximations can be proportional to the logarithm of the number of

HAMILTONIAN LANDSCAPES These beautiful images were created by Selim Tezel, M.Sc. ’02, using the software he designed for his Master’s project, “Traveling Salesman’s Sketchpad (TSSP 1.0).” The image to the left was a finalist at Manifest 2003, a juried digital visual arts festival, part of this spring’s international Boston Cyberarts Festival.

Designed and implemented to study the Euclidean Traveling Salesman Prob- lem, the process starts by placing a number of nodes (cities) on the plane that the salesman is to travel. Then an algorithm is chosen by the user (nearest neighbor, cheapest insertion, etc.) to compute an approximately optimal tour (approximate since the problem is NP-complete, i.e. not generally solvable in real time). The images rendered are isoperimetric and isopermutation regions associated with the chosen algorithms, demon- strating the constructed tours’ sensitiv- ity to initial conditions. A full descrip- tion of the process with a glorious im- age gallery and supporting docu- ments is at: http://www.cs.brown.edu/people/ stezel/webPageDesign/thesis.html

conduit! 3 rows and columns in an array. Finally, we have nanocomputers have not been fully devel- shown that highly efficient algorithms exist for oped, there is a role here to be played by programming some prototypical arrays when computer scientists in shaping the develop- h-hot addressing is used. ment of this emerging field. Conclusion Nanotechnology offers many interesting challenges. Although the building blocks for

that all major approaches to optimization be supported by high-level tools automating many COMET, OR HOW TO FIND of the tedious and complex aspects. Histori- THE RIGHT NEIGHBOR cally, most research has focused on con- straint and mathematical programming, and these approaches are now supported by Introduction a rich variety of modeling and programming tools. In contrast, neighborhood search, one Far from the public eye, modern of the oldest optimization techniques, has societies often rely on optimiza- been largely ignored until recently. This is a tion technology in order to func- serious gap in the repertoire of optimization tion effectively. From the airline tools, a limitation further exacerbated by the industry to courier services, from supply-chain management fact that solving optimization problems re- to manufacturing, from facility mains a highly ex-perimental endeavor: location to resource allocation, what will or will not work in practice is hard many important decisions are to predict. Proper software tools facilitate taken by optimization software this experimentation and often yield higher- every day. Even the entertain- quality solution techniques, since users are ment industry is no exception: more likely to try out various avenues. sport leagues, which are signifi- cant sources of revenue for radio At a conceptual level, neighborhood and television networks around search explores a graph whose nodes rep- the world, generate extremely resent solutions (or configurations) and challenging scheduling problems. arcs represent a transition from a solution to a neighboring solution. How to define Most optimization problems are Pascal Van Hentenryck this neighborhood graph and how to ex- challenging from both a computa- plore it effectively are fundamental issues tional and that have recently received considerable standpoint. Generally, they cannot be solved attention. in polynomial time and require ex- pertise in algorithmics, applied mathematics, and the application domain. Moreover, the resulting The distance between the software is often large and intricate, which makes it complicated to de- specification and the final sign, implement, and maintain. For instance, the apparently simple task program is considerable, of dispatching technicians to repair telephone lines requires about showing a serious lack in 150,000 lines of complex C++ code. expressiveness Since many optimization problems can be specified very concisely, this is very ironic. The distance between the Neighborhood search is the technique of specification and the final program is thus choice for a variety of fundamental applica- considerable, which indicates that software tions. For instance, at the time of writing, tools are seriously deficient in expressive- the best approach to the traveling tourna- ness and abstractions in this application area. ment problem, an abstraction of major- Because of the nature of optimization prob- league baseball scheduling, is a neighborhood- lems, no single approach is likely to be effective search method that significantly outperforms on all of them or even on all instances of a sin- constraint and mathematical programming. gle problem. It is thus of primary importance The same is true of many important problems,

conduit! 4 such as vehicle routing, frequency alloca- Once invariants are available, it becomes tion, and many resource allocation and natural to support the concept of differen- scheduling problems. Equally important tiable objects, a fundamental abstraction perhaps is the belief that hybrid algorithms for neighborhood search. Differentiable that combine several approaches in innova- objects maintain a number of properties tive ways are likely to produce the next im- (using invariants or special-purpose incre- provements in this area. Recent results in mental algorithms) and can be queried to routing and scheduling indicate the promise evaluate the effect of local moves on these of hybridization. Yet neighborhood-search properties. They are fundamental because algorithms are weakly supported in model- many neighborhood-search algorithms ing and programming tools. evaluate the effect of various moves before selecting the neighbor to visit. Two impor- The COMET project was initiated to address tant classes of differentiable objects are these needs. COMET, an object-oriented lan- constraints and functions. Constraints are guage supporting a constraint-based archi- a natural vehicle to model combinatorial tecture for neighborhood search, features optimization problems, as past research novel declarative and control abstractions. It has shown, but play a fundamentally nov- decreases the size of neighborhood-search el role in this architecture. Instead of programs significantly and enhances compo- pruning the search space as in systematic sitionaly, modularity, and reuse for this class search algorithms, constraints incremen- of applications. tally maintain properties, such as their vi- olation degree, and can be queried to The COMET Architecture evaluate the effect of transitions on these The neighborhood-search architecture sup- properties. ported by COMET (Figure 1) consists of a de- These first two layers, invariants and dif- clarative and a search component organized ferentiable objects, constitute the declara- in three layers. The kernel of the architec- tive component of the architecture. The ture is the concept of invariants over alge- third layer of the architecture is the braic, set, and graph expressions. Invariants search component, which aims at simpli- are expressed in terms of incremental vari- fying graph exploration. Neighborhood- ables and specify a relation that must be search algorithms often induce intricate maintained under assignments of new val- control flow, interleaving various applica- ues to its variables. For instance, the code tion aspects such as the neighborhood ex- fragment ploration, the heuristic, and the meta- inc{int} s(m) <- sum(i in 1..10) a[i]; heuristic. The control abstractions of COM- ET aim at separating these orthogonal fea- declares an incremental variable s of type tures in the source code, enhancing int (in a solver m) and an invariant speci- compositionality, modularity, and reuse of fying that s is always the summation of the search components. Events and first- a[1],...,a[10]. Each time a new value class closures connect these features con- is assigned to an element a[i], the value veniently by separating dynamic behav- of s is updated accordingly (in constant iors from their uses and when they must time). Note that the invariant specifies be used. the relation to be maintained incremen- Figure 2 depicts a simple COMET program tally, not how to update it. to solve the queens problem, placing n queens on an nn× chessboard so that no Search two queens attack each other. Line 5 de- clares an array of incremental variables Differentiable objects and queen[i] specifies the row of the queen in column i. Lines 8-14 specify the declarative component, i.e., the con- Invariants straints that must be satisfied by a solu- F un ts tion. The constraints in lines 9-11 specify cti rain that the queens cannot be on the same row ons nst Co and the same diagonals. They use the ubiquitous Alldifferent constraint, a differential object that arises in numerous resource-allocation applications. All the constraints are stored in a constraint system (also a constraint) that can be then queried Fig. 1. COMET’s constraint-based architecture in various ways. Constraint systems are one

conduit! 5 1. int n = 512; 2. range Size = 1..n; 3. LocalSolver m(); 4. UniformDistribution distr(Size);

5. inc{int} queen[i in Size] (m,Size) :=distr.get(); 6. int neg[i in Size] = -i; 7. int pos[i in Size] = i;

8. ConstraintSystem S(m); 9. S.post(new AllDifferent(queen)); 10. S.post(new AllDifferent(queen,neg)); 11. S.post(new AllDifferent(queen,pos)); 12. inc{set{int}} conflictSet(m); 13. conflictSet <- argMax(q in Size) S.violations(queen[q]); 14. m.close();

15. while (S.violationDegree()) 16. select(q in conflictSet) 17. selectMin(v in Size)(S.getAssignDelta(queen[q],v)) 18. queen[q] :=v;

Fig. 2. The queens problem in COMET of the main tools to enhance composition- The above example is of course very simple. ality and make it easy to add new con- What is interesting is that larger programs straints without changing the rest of the share the same structure and simplicity. For program. Line 13 is particularly interesting: instance, a job-shop scheduling program it specifies conflictSet, the set of queens written by Keith Schmidt (ScM ’01) about under attack. This set is specified by query- 5,000 lines of C++ long now takes less than 500 ing the constraint system and is automatical- lines of COMET for the same runtime efficiency. ly maintained by the COMET implementation whenever queens are moved. Lines 15-18 Implementation and Future specify the search, which consists of choosing Development a most violated queen and moving it to a new position in order to minimize its violations. The COMET system is being developed jointly Observe the disconnection between the prob- with Laurent Michel (PhD ’99), now an as- lem constraints and the search. Whenever sistant professor at the University of Con- line 18 is executed, all the constraints and necticut. It consists of an interpreter, a JIT invariants are automatically updated by , a runtime system, and many li- COMET, although they are physically separat- braries. The overall system is now close to ed in the source code. 200,000 lines of C++. The core, i.e., the com- It is interesting to discuss enhancing this piler and runtime system, is about 50,000 program for algorithm animation. Assume lines long, the libraries now becoming the that we have at our disposal an Anima- major part of the system. COMET has been tion class that provides a method evaluated on many applications. It has been shown to be competitive with special-pur- updatePosition(int q,int p) pose C++ programs on a variety of schedul- to update the row of the queen on column ing and resource allocation problems, while q to r. It suffices to add reducing the size of the code significantly. forall(q in Size) More importantly, perhaps, COMET was key whenever queen [q]@changes(int or,int nr) in helping us develop new neighborhood- animation.updateQueen(q,nr); search algorithms in car sequencing, ware- just between lines 14 and 15. The code features house location, and scheduling that have an event whenever that executes the enclosed solved open problems and/or outperformed closure whenever a queen is moved. Once existing algorithms. again, there is a complete textual separation Given the speed of progress in neighborhood between the animation code and the rest of the search, COMET is only a first step in a fasci- application. Note that solving the 2K queens nating area. Numerous open issues remain, problem takes about one second on a modern ranging from language design to runtime sup- PC. port and incremental algorithms necessary to

conduit! 6 accommodate a wide variety of new results exhibit these “exotic” constraints: just think (e.g., very large-scale neighborhoods and about all the safety regulations in the airline scatter search). Paramount among the lan- industry or, on a lighter note, the fact that guage-design issues is the need to preserve Duke must meet UNC on weeks 11 and 18 the compositionality, modularity, and effi- during the NCAA tournament, destroying ciency of COMET for applications with numer- the symmetry of the round-robin tourna- ous side-constraints gravitating around the ment. And we are not even talking about the core structure. Practical applications naturally constraints imposed by the NFL!

But I am getting ahead of myself now, for I HOW A SMART ELEVATOR wanted to tell you how it all began. We were both young then, new students in the Ph.D. CHANGED OUR WAY OF LIFE program, in fact. Guy Eddon, my partner in crime, was a struggling writer who had FOREVER made ends meet over the years writing Har- lequin novels while secretly pining for a bet- ter life. He hoped to find it in a town called Providence. For my part, my dot-com days of selling fake vomit over the Web had gone bust like so many other tinsel dreams of the web craze. I thought maybe I could make it as a pimp in Frank Woods’ escort service, but when business went dry, I decided to give Computer Science a try. In short, Guy and I were a couple of down-and-outers looking to better our fate. We had no reason to end up any better than all the other poor saps who entered the pro- gram, but we caught a break early on when one of the greats of those days, Ugur Cetint- emel, took us under his wing. Ugur began to teach us of the black arts he had devoted his life to mastering: Pervasive Computing, sometimes also called Ubiquitous Comput- ing. It was then that we caught our first glimpse of this netherworld, a place where the convergence of Moore’s Law and the vi- sion of Mark Weiser made it possible to Guy Eddon (l) and It seems hard to imagine now, but there was imagine embedding computation and com- Matt Lease a time before all the Gap ads, before the ap- munication into virtually any manufactured pearances on Oprah and the late-night talk object. In this world, the ubiquitous pres- shows, before the piles of fan mail and invi- ence of sensors and actuators enabled com- tations to posh parties had begun to arrive, puters to maintain a constant vigil a time before anyone knew we were any- monitoring human needs and the state of thing but a couple of graduate students with the world, continually seeking new opportu- a crazy idea. No one believes us now when nities to support human activities through we try to explain how something so simple appropriate physical and digital interven- revolutionized our way of life so completely, tion. But Grand Master Cetintemel had one but the world needs to understand how it all problem he hadn’ been able to solve: how really happened. At the time they told us it could this netherworld and the world we was impossible, that nothing like what we knew be brought together? How could all of envisioned had ever been done or could be the many interesting everyday artifacts and done, that we should put our youthful legacy devices from our world be enabled to dreams aside and face up to the reality of participate in this digital netherworld when what could never be. But like so many other they lacked the requisite computational dreamers, we persevered in spite of their abilities? criticism and the other challenges we met, for we knew if we could pull it off, if we could It was then that the idea took seed in us and do this thing they called impossible, it would began to grow. Legos had solved all of our be like nothing the world had seen before. problems when we were boys; could they

conduit! 7 save the day once again? It seemed so obvi- ize the elevator as a pervasive device. Not ous to us: Lego Mindstorms were cheap and only would the system be remotely operable could sense, actuate, compute, and communi- (one would be able to call an elevator in ad- cate—why not use them to sense and control vance or see if one had to hurry to catch it), everyday devices and revitalize them as new, but, by combining a location-aware badge pervasive artifacts? When we shared our system with an electronic appointment cal- idea with our friends and colleagues, we endar, maybe you would never have to call were universally condemned, belittled, and the elevator at all: the elevator would be able scoffed at, but who could blame them? You to anticipate your need for service. must remember that no one knew about the Although the Grand Master had his doubts, he consulted his brethren in the Dark Arts, and Overlord Stan Zdonik saw fit to provide no one dreamed that a simple us with the Lego Mindstorms we needed to undertake our mission. We toiled and slaved child’s toy could hold the key away in the dark recesses of the CIT, break- to such a breadth of problems ing from our labors only long enough to par- take in the weekly TGIF festivities and as NP-completeness, world consume sufficient Klondike bars to fuel our efforts. We faced many trials and tribula- hunger, and the six degrees of tions during the ordeal. When deploying our devices in the lobby, the evil ones, those who Kevin Bacon actually wanted to use the elevators, would see the exotic lights and wires of our devices Universal Turing-Lego Thesis then. Back in and constantly bombard us with nefarious those days “real” research required math, questions like, “Is the elevator working?” simulation, and independent thinking: no Yes, we would say, again and again, the ele- one dreamed that a simple child’s toy could vator is working, but then another minion of evil would approach us with the same ques- hold the key to such a breadth of problems as tion, thinking such incessant distraction NP-completeness, world hunger, and the six would prevent us from realizing our vision. degrees of Kevin Bacon. As we’ve said in the But all our engineering training was not for many interviews since then, we didn’t have naught, for we conceived a plan to beguile any idea ourselves of the revolution we were our assailants: we posted signs with craftily starting—all we wanted was to prove our- designed messages like “THE ELEVATOR selves to the Grand Master and earn an A in IS WORKING!!” And when they thought to CS295-1. ask us, “What are you guys doing anyway?”, We needed a way to test our idea, however, we handed them a flier and resumed our and it was then that we turned our gaze up- task. El Cid would have been proud of our on the CIT’s elevators. When we had first ar- determination. rived at Brown, we were amazed at the irony But the forces of evil were not to be assuaged so of the school’s Center for Information Tech- easily. One of the truly evil ones—you know, nology being saddled with a legacy elevator those who get off at the third floor—summoned system widely regarded as broken. Despite the night watch to harass us further: “It’s been the close physical proximity of the two eleva- reported that you guys are hacking the eleva- tors, they had been installed without any co- tor control system in order to bypass security ordinating logic—each had separate call protecting the basement!” To overcome as- buttons and ran independently of the other. saults, we obtained a scroll from Grand Master Even worse, the only visible indicators of the Cetintemel testifying to the righteousness of elevator’s position were in the lobby. Conse- our endeavor and threatening his great wrath quently, riders on other floors had no idea upon any who opposed us. With scroll in hand, which elevator to call for quickest service or our opponents quivered in fear and none could how long they might have to wait for an ele- stand against us. Our toil continued, day after vator to arrive. While we had no desire to day. There did come a time when we doubted, mess around with elevator internals (we when it seemed we lacked sufficient Legos to didn’t care how the elevator’s motors operat- accomplish our mission, but our prayers were ed and certainly didn’t want to do anything answered when his exalted mightiness Eugene requiring a new safety review of the installa- Charniak saw fit to bestow upon us a plethora tion), we figured we could augment the ele- of Mindstorms not to be used for the Robots vator with Lego Mindstorms not only to course until the following semester. And so we solve the problems above but also to revital- resumed our labors once more.

conduit! 8 At last the day of presenting arrived. We sis had been proven that we really began to plugged in all the wires, added the batteries, understand how deeply our work had and attached the most powerful device of all, changed the world, that nothing would ever actual duct tape, in great quantity. It be the same. For us, life has been a confused worked, and all who saw what we had ac- whirlwind of book signings and Turing complished marveled at its splendor. It was awards, extensions to Mount Rushmore, par- remarked that, after 15 years of no progress ties with Yanni, etc. I’ve come to realize that in fixing the elevator, a couple of graduate being famous, rich, and powerful is not al- students had come along and shown it was ways easy—I get hand cramps now and then simple enough to be solved using Legos. The from answering a mere fraction of the fan faculty swarmed around our invention, dis- mail—but I sleep better at night knowing cussing how all of the old textbooks needed we’ve made the world a better place, free to be rewritten, courses redesigned, and ac- from fear and envy and the injustices of the cepted practices revisited, and visitors from past, a place where universities understand all the lands came to see what we had done. the critical role of Legos in any serious re- Only then did Guy smile, for he knew then search agenda. that his days of Harlequin writing were be- “SmartElevator: Revitalizing A Legacy Device hind him and a bright new future was un- Through Inexpensive Augmentation” will be folding before us. presented at the 3rd International Workshop on Smart Appliances at ICDCS’03 in Providence, Looking back now, who could have guessed May 19th. Tickets should be purchased in ad- that would only be the beginning? It really vance as the event will, of course, be standing wasn’t until the Universal Turing-Lego The- room only.

Technical staffer Max Salvas adding 256MB of memory to Suzi's Maxbuilt system as well as upgrading her CPU from 800MHz to 1.2GHz. These enhancements were necessary to support the performance needed to produce this issue of conduit! During the past year the department replaced most of its desktops with Maxbuilts. During the summer of 2002, Max and Charles Williamson, the Student Hardware Technician, built more than 250 of these systems, bringing the depart- ment total to over 360. In addition to the generic Maxbuilt (1.5GHz CPU, 500MB memory, 40GB disk space, and GeForce3 graphics), Max has built several custom systems with high-performance dual processors and large amounts of memory. Max received a University Staff Bonus Award for his efforts and successes in the recent upgrade project.

conduit! 9 THE SECOND KANELLAKIS MEMORIAL LECTURE

The second annual Kanellakis Memorial Lecture was presented last December 9 by Prof. Christos Papadimitriou of the Uni- versity of California, Berkeley. In memory of our colleague Paris Kanellakis, who died Christos Papadimitriou with his family in an airplane crash in December 1995, these lec- surrounded by Kanellakis tures are held around Paris’s birthdate, December 3. Fellows. l to r: Yiannis Papadimitriou spoke on “Algorithmic Problems in the Internet,” dis- Tsochantaridis (Brown), cussing how the Internet, and particularly the World Wide Web, has Christos Kapoutsis (MIT), influenced research in theoretical computer science. Christos ex- Manolis Kamvysselis (MIT), pressed his belief that Paris, a person of broad intellectual pursuits Nikos Triandopoulos (Brown), who loved to tackle hard problems, would have enjoyed working on Manos Renieris (Brown), the sort of problems detailed in his talk. Olga Papaemmanouil We were especially pleased that in addition to the past and present (Brown), Aris Brown recipients of Kanellakis Fellowships, the holders of this year’s Anagnostopoulos (Brown), Kanellakis Fellowships at MIT were also present. Yiannis Vergados (Brown)

The meeting was kicked off by Bill Woods THE 30th IPP SYMPOSIUM of Sun Microsystems, who addressed INFORMATION & practical knowledge representation for effi- cient support of information access and in- KNOWLEDGE MANAGEMENT formation sharing within an organization. One of the fundamental problems in this On November 14, 2002, the Department context is apparent to anyone who has used hosted the 30th IPP Symposium on “In- information-retrieval tools such as search formation and Knowledge Manage- engines: not everybody uses the same words ment”, a topic at the frontier of today’s and phrases to express a certain fact, con- computer science research: How can cept or idea, as in “calling a spade a bloody computers support the creation, preser- shovel”. This is called the vocabulary mis- vation, and dissemination of information match problem. To address this problem, and knowledge? Three key issues came Bill has devised a method that uses a con- up time and again in the talks and dis- cept taxonomy to match a search request cussions. First, how can we build search with a document on the concept level (not engines and retrieval systems that can simply by comparing keywords, as most better deal with the ambiguities and in- search engines do). Thus the system can un- herent vagueness of natural language? derstand that someone looking for informa- Second, how can we extract useful infor- tion on “dogs” may be interested in a Thomas Hofmann mation or knowledge from repositories document on “German shepherds” but is like the World Wide Web? And third, how less likely to want information on “under- can we make multimedia documents such dogs”. as video and audio searchable?

conduit! 10 Symposium speakers l to r: Andrew McCallum, UMass Amherst; Bhiksha Raj, MERL; Susan Dumais, Microsoft Research; Bill Woods, Sun; Alex Cozzi, IBM Almaden, center; Pedro Moreno, Hewlett Packard; John Smith, IBM Watson; and host Thomas Hofmann Bhiksha Raj and Peter Wolf of MERL standard database record. With this tech- Mitsubishi Research Lab in Cambridge nique, FlipDog has automatically collected presented SpokenQuery, a system that pro- and processed more than 200,000 job ads— vides a spoken interface to an information- and keeps on going! At its core, this impres- retrieval system. Ever want to google sive system is powered by sophisticated ma- through your cell phone or PDA? Don’t like chine-learning algorithms using statistical the foldaway keyboards, and don’t like draw- models that improve on a standard tech- ing your query in glyphs either? Well, then, nique known as hidden Markov models—a SpokenQuery is what you (will) need. Just wonderful example of how progress in mod- speak what you want to know into a micro- eling and theory can make a real difference phone! Its automatic speech recognizer is op- in practical applications. timized to get at least the essential parts of your query right, i.e. those important key- Susan Dumais of Microsoft Research in words that really matter. It does this by cou- Richmond, a renowned research veteran, pling the speech recognizer with the presented research by her and her col- document index, since it is the index that is ul- leagues on question-answering retrieval us- timately used to retrieve relevant information. ing the World Wide Web as a knowledge source. This addresses one of the longstand- The next presentation was given by a re- ing dreams of artificial intelligence: how can searcher with a stellar reputation in both we build a computer that can answer any the academic and commercial world, An- question we mere humans may ask? Well, drew McCallum of UMass Amherst, why not use the hidden knowledge treasures former VP of R&D at WhizBang Labs. His of the Web to do it! Of course, this will just be research is driven by one key question: how a collective echo of everyone who has put can we transform the “Wild Web” into a content on the Web, but maybe this is suffi- knowledge base? To demonstrate the feasi- cient for many questions people have. So bility and the commercial value of this idea, how, in a nutshell, does their system, called Andrew and his coworkers at WhizBang AskMSR, work? First, the question is trans- Labs developed the FlipDog job search en- formed into a number of answer templates. gine (www.flipdog.com). Using innovative You ask, “Where is the Louvre Museum lo- information-extraction tools, FlipDog con- cated?” and AskMSR looks for Web pages stantly spiders the Web in search for its containing phrases like “the Louvre Muse- prey: job ads. Once a job ad is found, relevant um is” and inspects the context to the right. information such as company name, job The collected snippets are then filtered and type, department, level, location, etc. is ex- finally tiled together based on redundancy tracted from the HTML file and mapped to a and overlap to create the answer. Does it

conduit! 11 work, you ask? Amazing- software like this to search through ly well: “Who is Bill Gates the webcast of your favorite CS course married to? AskMSR’s for keywords like “inheritance” and answer is “Microsoft”. “polymorphism”, but what would be the fun in doing that? In their talk on Model- Andy was delighted to Based Retrieval and Mul- In the last talk of the symposium, report that his student Leah timedia Information and Biosurveil- Pedro Moreno of Hewlett Packard lance, Chung-Sheng Li and John Research Labs presented a truly fas- Pearlman, a junior and R. Smith presented some recent re- cinating system for audio retrieval search projects at IBM Watson Labs called SpeechBot. Collecting hundreds Sc.B candidate has been on multimedia retrieval and data min- of gigabytes of data from radio news ing. Among the many highlights of feeds, SpeechBot automatically tran- awarded a Microsoft Schol- their talk was their account of a sys- scribes the speech to text and makes tem for biosurveillance, a prototype of the information searchable by building arship for next year. This which was deployed at IBM Watson. an index. SpeechBot (speechbot.re- scholarship covers full The system aims at automatically de- search.compaq.com) currently indexes tecting early warning signs of possible an impressive 16,000 hours of radio— tuition and includes a 12- outbreaks of epidemic diseases, that’s almost three years of listening whether of “natural” or “less natural” day and night! The main challenge is week salaried summer causes. The system collects data on that the transcripts generated by personal activities like phone calls, speech-recognition software rarely ex- internship at Microsoft. employees’ arrival and departure actly match what was spoken, because times, Internet access patterns, etc. different people speak at different Our thanks to Microsoft to mine for irregularities and alarm- speeds and with different degrees of ing trends in people’s behavior. The clarity. However, meaningful words and congratulations to idea is that a system like this can be are recognized with high accuracy and Leah! an early warning system to give pre- even when a word is missed, it will cious lead time compared to tradi- probably be recognized when it is spo------8<------tional channels that rely on sources ken somewhere else in the program. In like medical reports. addition, a number of innovative tech- Doctoral student Manos niques have been implemented to Existing video libraries are growing counteract the unreliability of today’s rapidly. Soon we will be facing the Renieris received rather automatic speech-recognition systems, problem of finding relevant informa- for example, the use of sub-word units, back-handed kudos as a tion in the giga- or terabytes of data so-called ‘particles’, to index spoken stored. Alex Cozzi of IBM Almaden documents. For example, the word result of his failed applica- Research described CueVideo, a pro- “pescatarian” is unlikely to be recog- totype system developed at IBM that nized by any speech-recognition sys- tion to the ICSE ’03 Doc- can create browsable video summa- tem, but the particle “tarian” may be ries and automatically index video toral Symposium. Said his useful as an index particle that can be footage based on an audio transcript. matched with queries like “vegetarian”. Say you own (or illegally copied...oops) reviewer, “Your application a movie like Pulp Fiction and you Now let’s hope all this will improve the was reviewed as something want to find the scene on “foot mas- state of the art in intelligent informa- sage” or the “catch-up/ketchup” joke. tion retrieval. If you AskJeeves, “The perhaps best described as CueVideo might help you get right to meaning of life is to increase fitness”. the point. Of course, you can also use Am I missing something here? ‘Has no real need of the

Dr. Tom Tullis, Sr. VP of Symposium.’ The applica- Human Interface De- sign at the Fidelity Cen- tion showed that you are ter for Applied Technology, joined making excellent progress Michael Black’s CS32 Intro. to Software Engi- under excellent supervi- neering class to lec- ture on “Effective User sion, and with considerable Interface Design”. Tom founded and runs the success in the field already. Usability Lab in Boston. However, we want you to Fidelity Investments is a CS Industrial Partner. know that except for being

conduit! 12 LETTERS FROM ALUMNI AND FRIENDS ‘overqualified,’ you would have been selected. We ED BIELAWA, ScB & ScM think about how many ways you can ask for the phone number of the CIT! And would like to offer a sort of ’95.5 that’s assuming callers respond to After eight years and six companies prompts in the correct format—but consolation prize: Your many callers speak as freely as riding the Internet startup wave (with very short abstract will one buyout but otherwise a lot of stock though the system could understand certificate “wallpaper”), I’m currently natural language. Fortunately, the appear in the ICSE Pro- working on the other side of software de- number of calls to directory assistance velopment: I’m a product manager at is somewhat limited, but we still need ceedings, and we would Phonetic Systems in Bedford, MA, technology to handle free-form natural- where I define the requirements of our language responses (Eugene’s CS241 value your attendance at speech-recognition platform and man- has come in handy!). Even after the da- age its integration with partners such tabase is normalized and callers are the Symposium to partici- as Oracle, AT&T, and IBM. My focus is directed to correctly formatted re- twofold: on the enterprise side, to in- sponses, speech recognition on large- pate as an informal mentor. crease our market penetration by part- metro-area databases such as New nering with interactive voice-response York or Los Angeles challenges the You will have free admis- computational limits of existing hard- systems (Brown is using our system to sion to the full ICSE con- voice-enable the campus directory), and ware—and searches must be done in on the telco side, to build and deploy the real time while callers are holding on ference just as those first fully automated national directory the line. assistance “411” system. Our directory assistance automation accepted for the Sympo- Contrary to popular belief, directory as- platform is currently deployed by sistance is not automated today. Even Fonecta, the largest phone company sium do.” though a computer prompts callers for in Finland, and is in various trials their listings, live operators listen to the stages with every major US phone ------8<------recordings and manually enter the re- company. It’s been fun and challeng- quested listings into a computer. While ing to learn about a new technology We’re pleased to announce speech recognition on small databases is and its applications, especially from almost a commodity nowadays, speech the other side of the fence, where I get that we’ve received a gift of recognition on the entire US white pag- to build new partnerships, drive re- es (150M+ listings) is a very difficult quirements to R&D, and manage 15 Tablet PCs and associ- technical integrations with partners. task. First, the database requires signif- ated software from CS icant normalization, since listings have Many of my co-workers think our been entered in many different ways by CEO is difficult to work for since he Industrial Partner many different operators over the years. frequently calls for status updates For example, names and roads can be (and to check if we’re in the office); but Microsoft Research. abbreviated in many different ways, and it’s not a problem for me since I was callers never ask for listings the way prepared to work for a taskmaster by These will be set up in a they appear in the phone book—just Andy as a CS15 Head TA! I’ve also ap- preciated the support and networking Tablet PC lab for under- of Brown CS faculty and alumni for my Internet startups. I’ll graduate education, which probably return to development at some point, but for now, I enjoy will be used by Andy van product management, and I get my coding fix writing stock and Dam’s ReMarkable Texts sports analysis programs and play programs for my son, Eddy project and Tom Doepp- (Brown CS ’24?). Eddy already likes to play on the computer, and ner’s Electronic Student now that he can walk and talk, Notebook project. Staff programming is just around the corner! Email is welcome at members Loring Holden, [email protected].

conduit! 13 CHRISTOPHER a relationship or assumed role may evolve into something unanticipated. ELAM, Public [email protected] Policy ’99 CINDY GRIMM, PhD ’96 Tim Miller, and Bob While at Brown I spent much time in CS classes Bill Smart (PhD ’02) and I got married Zeleznik and undergrad and as part of the Graphics Group. last summer—twice, once in Scotland Yotam Gingold are partici- Since school I have pursued my inter- and once in Oregon. We both joined the ests in dance, receiving an MFA in faculty at Washington Univ. in St. pating in the former; grad Dance at NY Tisch in 2000. I have con- Louis two years ago. We’ve found tinued to develop my own company, Wash. U. to be very like Brown Univer- students Roberto Almanza the Misnomer Dance Theater, which sity, especially the Computer Science was founded in 1998, in my senior department. The students are bright, and Qiang Ye and ugrads year at Brown. I often argue that ob- dedicated, and a pleasure to work with. ject-oriented programming has actu- We had a very busy summer, since in Albert Huang, Ulises ally become a useful metaphor for addition to getting married we had our Ortiz, Luke Peng, and some of my choreographic techniques. first exhibition of “Lewis, the Robotic “Computer science and dance are Photographer” at SIGGRAPH. Lewis Chris Wells are participat- closely knit in my mind” is what I say is a red, four-foot tall robot that takes to people wondering about this cou- pictures of people at parties. The ing in the latter.The tablets pling. Since its formation in 1998, project has been (rather unexpectedly) Misnomer Dance Theater has found very popular with the press, including will make their debut in tenderness, humor, and absurdity in the BBC and CNN. The students think people’s efforts to relate to one anoth- this is great because they’re working Doeppner’s networking er. As Artistic Director, I challenge on a project that their mothers can traditional representations of ro- read about in the local paper! course (CS168) at the end mance, compassion, and closeness by creating moments in which people Last summer I also wrote an NSF CA- of this spring semester and connect in the most unusual ways. REER proposal, which, I am pleased to Synthesizing my extensive training in find out, has been funded. The propos- will appear again in his both Balinese and modern dance, I al’s title is “A Composition System for operating systems course hope to establish a cultural exchange Computer Graphics”. The idea is to that happens in the body itself. The combine my background in art with (CS167) and van Dam’s tension and sculptural definition of my interests in computer graphics to Balinese dance is integrated into the create tools that will help the “average graphics course (CS123) fabric of my broken-flow movement. I user” make more comprehensible im- chose “Misnomer,” the misuse of one ages. Artists learn a large number of this fall. We very much word for another, to describe the way techniques for making images easier to understand; can we find ways to appreciate this support make these techniques available to non-artists? from Microsoft Research. JOHN MARTIN, ScM ’83 ------8<------Hi Andy: It has been a long time, and The 03-04 academic year is you’ve had thousands of students, so you may not remember me. I received the department's 25th! my ScM in 1983 and then went to work We're planning a special Above: Chris and Andros Zins-Browne for your grad school officemate, Dick ’03; below: Jocelyn Tobias and Chris Wexelblat, at two companies. distinguished lecture series I read your address in the latest issue of conduit! last night, and it struck to honor this event as well me that it is time to thank you for your contributions to my education. Even as a number of other festiv- though I was only at Brown for a year, ities, including a celebra- the time was very concentrated and in- tense. I was used to working hard be- tion on May 27, 2004, the fore I got to Brown, yet you still managed to challenge us in a new way. Thursday before I remember sleeping on the concrete

conduit! 14 floor of the Cause Lab, face down, in between tracking down bugs. What we built as a project in your course was comparable to what Yahoo Maps does right now, and I Commencement. Please am so proud that my team built that in a semester project. keep this date open! On a personal level, I am still very good friends with people I met in ------8<------your graphics class 20 years ago. As you can see from Don I have been working at NYNEX/ Bell Atlantic/Verizon for the past Stanford’s CS002 spring 12 years. If I look out of my build- weekend Hawaiian shirt ing, I can see the building where Blue Sky Studios created the movie “Ice Age”. I imagine that there are some of your students there right now creating the next big animated film. It seems like only yesterday that I visited Brown CS for the first time, and you spent a lot of time with a few of us prospective students in your office. I was sold at that moment, and going to Brown was one of the smartest things I’ve done in my life. Thank you. John Martin. [email protected].

PYGMY SEAHORSE The photo is of a pygmy seahorse (Hippocampus barg- ibanti), taken last August off the island of New Ireland, Pap- ua New Guinea at a depth of 60 feet. Such extravaganza, they seem to seahorses are quite small (this one is less than a quarter-inch long) and incredibly hard to find— have raised the bar on this they look exactly like the coral they live on. They event! were discovered only after someone collected some coral, put it in an aquarium, and then (fi- nally!) noticed the tiny seahorses, probably ------8<------dead by this time. Knowing that the coral at this particular location had some pygmy seahorses Renovations to the CIT 3rd on it, it took my two dive guides and me 10 min- utes to find them. They were once thought to be floor will begin in June so rare, but, now that divers know to look, they are turning up at numerous sites in the South Pacific. CS can expand into it by September. This means we For the camera literate, this was taken using Fuji Velvia can admit more graduate film and a Nikon F4 in a Nexus housing at an aper- students and have space for ture of f22 at 1/250 second with two large Ikelite more faculty and visitors. strobes through a 105mm lens at closest focus ------8<------with a closeup adapter in- creasing the magnification CS is on the main Brown by 90%. web page again; that’s grad Tom’s newly updated web page with many recent under- water photos is at: student Dan Keefe in the http://www.cs.brown.edu/peo- ple/twd/fish/start.htm. Cave! www.brown.edu

conduit! 15 NEWS FROM KEITH ADAMS ’00

Much has changed since my graduation outs. After all, we would have TAed some of only three years ago. When I packed up my these people back at Brown. Providence apartment to join a small virtu- VMware, the company, is increasingly hard alization software company out west, I left to describe concisely. We’re a system soft- behind a Brown very different than the one ware company that began as an outgrowth present students inhabit, and I was enter- of Mendel Rosenblum’s virtual-machine re- ing a very different world than the one search at Stanford. For those who were nap- greeting the class of 2003. ping that day in 167, in the ’60s and ’70s virtual machines enabled mainframe users 2000 was the Internet Boom’s swansong to make huge IBM boxes look like multiple, year. Every CS graduate who cared to could smaller machines. Even though computers paper a bedroom wall with job offers. are cheaper today than in the mainframe’s Brown’s Sun lab actually contained Sun heyday, virtual machines remain useful, computers. A departmental Berlin Wall perhaps more so than ever. Lest I lapse into split the systems lab in two, and the inhab- starry-eyed marketspeak, I’ll refer those cu- itants of Syslabs A and B regarded each oth- rious about the applications of VMware’;s er with suspicion and distrust. Many CS products to our web page (www.vm- courses were crowded with students who ware.com). seemed to have a passing interest in compu- My three years here have been deeply satis- tation but a profound interest in making six fying. VMware has been successful during figures while hacking javascript. And those very difficult times, growing from a rock’em, of you expecting to graduate in 2003 were sock’em startup with 80 employees to a freshmen, meaning that the modicum of more stable, though small, company of 300. face recognition I still enjoy on recruiting trips We’ve gone from shipping a single product back to the CS department will soon be gone. to shipping three major products (Worksta- tion, GSX Server, and ESX Server). Matt and I both played important ...Brown consistently produces roles in the bringing the world’s some of the best computer first multiprocessor virtual-ma- chine product to market. Across the science graduates in the world. I board, we have been entrusted with much more responsibility than knew abstractly in 2000 that I’d larger companies would ever con- sider giving a pair of young punks gotten a great education at straight out of school. Brown, but working at VMware has We’ve also been joined by fellow Brownistas Paula Robin (’01) and truly driven the point home Rob Manchester (’??), with hope for a still unfinalized handful of folks VMware will keep sending me back, to follow this summer. And we hope though, because Brown consistently pro- to see more in the future, for Brown CS and duces some of the best computer science VMware, like chocolate and peanut butter, are graduates in the world. I knew abstractly in two great tastes that taste great together. 2000 that I’d gotten a great education at Of course, a Brown degree is no guarantee Brown, but working at VMware has truly that a person is bright, talented, or motivat- driven the point home. ed. My year certainly contained its fair share of unredeemable goofballs; I’m sure When Matt Eccleston (’00) and I showed up you think the same is true of your year. at VMware’s anonymous office space in the However, having been involved in recruit- summer of 2000, we were among the first ing from other ostensibly top-tier CS engineers employed there without graduate schools, I think that Brown is weirdly good degrees. For most new college grads, being at nurturing those with the potential into both so new and so junior would have been successful practicing computer scientists. intimidating. This pair of veteran CS169 I’m not sure why this is. So much is atypical TAs, though, was uncowed by the mix of about the whole Brown CS experience that professors, PhDs and PhD program drop- it is difficult to match causes to effects. Is it

conduit! 16 the undergrad involvement in TAing and re- have been poorer, or at least less enjoyable, search? The faculty’s admirable focus on un- anywhere else. dergrad education, rare at a research It’s possible to go overboard, of course. Some university? These certainly play a part. But material is hard to teach in an applied style, I’ll go out on a limb and claim that, for at and 152 rocked, though I didn’t write a line least some students, it’s mostly the coding. of code for it. Now, there are those who suspect that any Thus far, I must rate my transition from programming in an academic class is sub Brown CS student to productive member of rosa job training. I understand this fear; I society a success. Many, many Brown CS recall with a shudder those javascript jock- folks end up in the San Francisco Bay Area, eys who flooded my classes a few short so the social blow of relocation has been years. I am not advocating that we turn greatly cushioned. I considered listing all Brown into a trade school. Rather, I think the Brown CS people I’m in contact with out that for people of a certain psychological here, but it runs to 40 or so. It seems at times makeup, coding is important purely as an to be a veritable CS department-in-exile. educational tool. When I was a student, the When I want to wax nostalgic about elevator little blip of seratonin that came from keys, prank-calling the Sunlab consultant, translating lecture slides into a functional getting wings and watching “Shaft”, or lump of software helped cement those lec- watching those three blinking red lights ture slides in my brain. There are precious from the pink couches on the fifth floor at few other top-rank CS schools that offer the night, there’s always a fellow Browniac opportunity to learn by doing as deeply and nearby, ready to hold my hand down memory broadly as Brown, and my education would lane. [email protected].

RARE TROPICAL VISITOR From January 9 until March 14, seven beautiful Zebina Hairstreak butterflies mysteriously ap- peared one at a time in Suzi Howe’s office! There are no flowering plants in the office, so Suzi was feeding them sugar water and deposited a couple next door in Trina Avery’s office, which happened to have some plants in bloom. Each survived from three to five days. From what we have gleaned from on-line sources, our butterflies were all females, their grey wings outlined in black with two additional orange, black and white lines defining the fore and hind wings and creating a streak (hence the name). Orange eyespots with black dots, the better to confuse predators, were located on the hind wing, where the wing shape changed to an elegant bracket curve with delicate white-tipped appendages trailing behind. These eyespots were on both sides of the wings and dramatically visible when the wings opened flat. When the sun shone directly on these little guys, the edges of their wings glowed a neon blue...we were definitely smitten! According to Butterflies of North America, this little-fingernail-sized visitor is considered a ‘rare tropical stray.’ It ranges from Colombia to Mexico and only three such strays have been re- corded in southern Texas, which makes our Rhode Island visitation all the more thrilling. Well, maybe not for everyone—when astaffer Fran Palazzo saw one she said, “That’s no butterfly, that’s a moth and it’s going to eat my sweater...where’s the fly swatter?” After Suzi emailed an emergency posting asking for a lepidopterist in the department to help solve the mys- tery, Shriram Krishnamurthi noted that he and his wife had just returned from a trip to Mexico over winter break and the hairstreaks might have flown out of their luggage...he admits this is a ‘wild speculation’ but sug- gested we should give a small award to the person who posits the most extravagant theory—a free annual subscription to conduit! (runner-up gets two!)...

conduit! 17 LIDDY SHRIVER ’90

Elizabeth (Liddy) Shriver’s father Bruce, a pi- which will hinder her great passion, cycling. oneer in computer architectures and micropro- Before this happens, she plans to take several gramming in particular, met Andy van Dam in long bike rides as part of a campaign to raise the ’60s. Since they have been friends so long, public awareness of sarcoma, especially to the critical lack of young adult participation in Andy was delighted when Liddy came to clinical trials. Because most Ewing’s sarcoma Brown. She worked under Jeff Vitter and re- research has been done on children, the results ceived her M.Sc. in 1990; her thesis topic was do not always apply to adults. This lack has re- “Optimal Disk I/O with Parallel Block sulted in a higher fatality rate for adults. “I fall Transfer.” After earning her doctor- into the category of young adults with cancer,” ate at NYU, she went to work at Bell says Liddy, “which means there isn’t much da- Labs. ta available to help my oncologist know what treatments might work for me.” Liddy’s health was normal until nine years ago when she became aware of On April 10-13, Liddy, 36, joined friends and a burning pain in her foot. After years family members for their first Team Sarcoma of this severe pain, a tumor was found bike ride at Cycle Zydeco, the second annual behind her knee, which a biopsy four-day, 200-mile southern Louisiana bike showed to be Ewing’s sarcoma/PNET tour. In late June and early July, Team Sarco- (primitive neuroectodermal tumor). ma will undertake a bike tour in Denmark. To Says Liddy, “I had chemotherapy, sur- join the team or for more information, please gery, radiation, and more chemother- see the following websites or contact Liddy at apy, but my cancer has spread to my [email protected]. lungs. So far, the nodules in my lungs www.theshrivers.us/la_bike_our.htm haven’t responded to treatment.” As www.theshrivers.us/danish_bike_tour.htm Practicing for the the disease progresses, Liddy will Denmark tour have increased difficulty breathing, www.cyclezydeco.com

cameras (right) and infrared illumination to detect and track reflective markers placed MOTION CAPTURE on the body of a subject (this is the same system used by Hollywood to transfer the mo- tion of human actors to animated characters). The goal of the human motion project at Brown is to develop al- Using the six cameras and special-purpose hard- gorithms that can “capture”, or ware, we collect the motion of various subjects estimate, the motion of people in performing a variety of tasks. Machine-learning video sequences. This problem is techniques are then used to model how humans challenging due to the variabili- move. ty of human appear- ance, the complexity of the human body and its motion, and the loss of three-dimensional in- Michael Black formation when a 3D scene is projected onto 2D images. If we can solve the problem there are applications as var- ied as animation, medical diagnosis, and human-computer interaction. To study this problem we are building a one-of-a-kind motion-capture facility at Brown.

The first component of this facility is a commercial Vicon motion-cap- ture system. This system uses six Vicon cameras with ring of infrared LEDs

conduit! 18 Capture session. Note reflective markers on the subject This system, however, is tion provided by the Vicon system. This will Temporary not fully general since it provide a state-of-the-art facility that can be used motion-capture requires that people to evaluate and compare current human motion lab wear special markers. estimation algorithms and to provide new types We would like a marker- of training data for machine-learning approaches less system that “understands” something about to human motion analysis. how people appear in images. To that end we This work is supported by the DARPA HumanID have installed a second video-based capture sys- project. The Vicon system was also partically tem that can save the video data from four high- funded by the NSF IGERT program and is shared speed progressive-scan cameras. We are in the with the VENLab in the Department of Cognitive process of synchronizing these two systems so and Linguistic Sciences. that we have video sequences of people from mul- tiple views along with the “ground truth” 3D mo-

[email protected]

MICHAEL BLACK. Michael gave a num- (IVIMEDS.org), an international collaboration ber of invited talks on brain-machine inter- of more than fifty leading medical schools and faces, including the keynote talk at universities worldwide including Brown’s EURON, the European Robotics Research School of Medicine. The Technology Group is Network annual meeting in Lisbon; a plena- involved in the development and evaluation of ry talk at the Ohio State Mathematical Bio- the reusable learning objects (RLOs), instruc- sciences Institute workshop on neural tional software, and other e-learning aspects coding; and a plenary talk at the First Inter- of the IVIMEDS project. national IEEE EMBS Conference on Neural Engineering in Capri. After many years of waiting he had a patent finally issued on work done at Xerox PARC with Allan Jep- son from the University of Toronto (“Appa- UGUR CETINTEMEL. Ugur served on ratus and method for identifying and the program committee of the Interna- tracking objects with view-based represen- tional Conference on Distributed Com- tations”). Siemens Corporate Research gave puting Systems, held in Providence this a generous $25,000 gift to support his re- May. He also worked on the organization search on human motion analysis in video of the Workshop on Internet Applications sequences. to be held in San Jose later this year. Ugur gave talks at the Symposium on Re- liable Distributed Systems in Osaka and the Conference on Innovative Data Sys- tems Research. This semester he started ROGER BLUMBERG. Roger has been teaching his new course, Networked In- named to the Core Technology Group of the In- formation Systems (CS 138), a hands-on ternational Virtual Medical School Project introduction to fundamental principles

conduit! 19 and practice underlying today’s networked never seen before. He found it mildly ironic information systems. that he had to move from Texas to Rhode Is- land to get a chance to cross the border. In the fall, Shriram picked up two more award paper nominations, one each from Au- AMY GREENWALD. Amy has received tomated Software Engineering and Founda- one of this year’s Richard B. Salomon Faculty tions of Software Engineering. He was Research Awards, established to support re- especially thrilled that both papers had a search projects of exceptional merit. Amy’s Brown undergrad, Harry Li, as a co-author. project will compare human decision-mak- Harry’s now in the PhD program at UT Aus- ing processes about correlated equilibria tin. Shriram also served on the program with the reasoning of computational agents. committees for Technologies for XML, Lightweight Lan- guages 2 and Practical Aspects of Declara- tive Languages. SHRIRAM KRISHNAMURTHI. Shriram January was Shriram’s month for invited spent part of winter break in Mexico. His talks. His Mexico visit was instigated by an highlight was visiting Teotihuac’an, which invitation, to him and three book co-authors, he had read about for nearly 20 years but to run a programming languages summit at the Universidad Nacional Autónoma de México. The next week he gave a keynote talk at Practical Aspects of Declarative Lan- guages, and the week after that he spoke in the Distinguished Seminar Series at IBM Watson. He’s very glad nobody’s inviting him to anything else right now.

DAVID LAIDLAW. In April David was awarded the prestigious Henry Merritt Wriston Fellowship for ’03-’04 in recognition of his innovative Brown/RISD course with Fritz Drury of RISD’s Illustration Depart- ment. CS and RISD students work together in teams to develop new ideas for the visual presentation of fluid-flow data, using tradi- tional ‘crits’ to evaluate and refine their de- signs. David is now an associate editor of IEEE Transactions on Visualization and Computer Graphics.

FRANCO PREPARATA. Franco recently spent a few weeks at the National Univer- sity of Singapore as the Qwan Im Thong vis- iting professor of Computer Science. He was Dan Keefe (drawing) and Cullen Jackson (behind recently appointed to the Scientific Board of camera) working on a user study investigating how the Istituto per la Scienza e la Tecnologia of graphic designers and illustrators critique and create scientific visualizations. Here Dan Keefe is the National Research Council of Italy and to creating a new data-driven scientific visualization the International Advisory Panel of the based on his analysis of other visualization Graduate School Integrative Sciences of the techniques previously viewed National University of Singapore.

conduit! 20 JOHN SAVAGE. John finished his third face between operations re-search and year as an officer of the Brown faculty and computer science. This prize, given annually a member of the Faculty Executive Com- since 1986, recognizes Pascal’s many contribu- mittee and a year as a member of the Aca- tions to constraint programming and its inte- demic Priorities Committee, a new gration into operations research. Pascal also committee introduced as part of a major gave invited talks or tutorials at the annual overhaul of faculty governance spearhead- INFORMS meeting in San José, at the IN- ed by the Task Force on Faculty Gover- MA school on large-scale optimization in nance, which John chaired. The Task Minnesota, at the Optimization Days in Force, which began its work about a year Canada, and at the New England Sympo- ago and finished in early April, introduced sium on Programming Languages and changes to increase collaboration between Systems. He was also program chair of the the faculty and the administration and re- International Conference on Constraint duce the number of committees from 44 to Programming, which attracted a record about 25 and the number of faculty slots number of submissions. His brother and from 237 to about 135. Last summer John father-in-law made a surprise visit on finished eleven years of service on the vis- March 8 for his 40th birthday, coming all iting committee for the Department of the way from Belgium. Electrical Engineering and Computer Sci- ence at MIT, his alma mater.

EILEEN VOTE. Eileen, a postdoc working with David Laidlaw, has been awarded one DON STANFORD. Don has been busy of this year’s Richard B. Salomon Faculty teaching CS002, which has a record 317 Research Awards for research projects students enrolled! In addition, he intro- deemed of exceptional merit. She will duced GTECH to a legal form of gaming work on a three-dimensional reconstruc- for the Internet, which will let them create tion of the water system at Qumran, the a product that appeals to Internet-centric find site of the Dead Sea Scrolls. A digital players. reconstruction of the water system will help derive a chronology for the archaeolo-

ELI UPFAL. Eli was an invit- ed speaker in a workshop on Topics in Computer Communica- tion and Networks at the Isaac Newton Institute for Mathemat- ical Sciences, Cambridge, UK, and served on the Program Com- mittee of WEA 2003, the Second International Workshop on Ex- perimental and Efficient Algo- rithms, held in May in Ascona, Switzerland.

PASCAL l to r: Cullen Jackson, RISD Professor Fritz Drury, VAN HENTENRYCK David Laidlaw, Daniel Acevedo and Dan Keefe meet in the atrium to discuss the critical Pascal was awarded the 2002 features of data-driven scientific visualizations INFORMS ICS prize for re- and how to combine them into perceptually search excellence at the inter- optimal images

conduit! 21 gy of ancient Qumran, thus providing a accident four years ago is being used by medical context for insight and discovery about the students to examine medical and ethical issues Dead Sea Scrolls. in helping people recover from brain injuries. His work as editor of the Faculty Bulletin allows faculty to write about improving teaching and research at Brown. He uses his office for reading and research most PETER WEGNER. Peter’s article in the weekdays, appreciates the department’s April 2003 CACM examines the impact of thoughtfulness in supporting his retire- Turing machines on mathematics and the ment and is working with Lew Lipsitt on theory of computation. His article in the Brown’s upcoming retirement community. forthcoming book on the 90th anniversary of Turing’s birth in 1912 examines the im- pact of Turing’s work on current computer science. A film by the Brown Medical School about Peter’s remarkable recovery from his serious

CHARNIAK UNPLUGGED

A few months ago CNN did a piece of publicizing this bit of web trivia, on the 25th anniversary of the smi- that it had taken off much more than ley “:-)”, now the ubiquitous indica- anyone expected, and that he was glad tor of email emotional state. I was things were dying back down so he surprised to see a colleague of could get some work done. mine, at CMU, ex- plaining that, as far as he or any of A few days later I mentioned all this at his colleagues knew, he was the a lab meeting, and one of my graduate first person to use a smiley, and students, Heidi Fox, said that SHE that he did so in an email sent 25 knew the fellow who invented the use years ago. I have known Scott for of the “@” in email! I quickly realized somewhat longer than that, as we that this was surely not an idle boast. overlapped as MIT grad students Before she became a full-time gradu- in the early ’70s, and since his field ate student Heidi worked at BBN, the is also artificial intelligence we company that had had a major role in have frequently met at conferences the construction of the so-callet “AR- Eugene Charniak ever since. However, I had had no idea PANet,” the predecessor of the Inter- that he was so distinguished! net and ARPA’s (DOD’s Advanced Research Project Agency) major claim The CNN interview was pretty funny. to popular relevance. At any rate, it They asked him if he ever got any fi- was clear that I had been one-upped, nancial reward for this major inven- since as even Scott Fahlman would tion. He responded that he had not, and have to admit, the “@” is more critical to the follow-up question of whether he in current daily life than the “:-)”. wished he had, he responded with good humor that if he got money for every Many conduit!s ago I explained how use of the smiley, people would proba- it came to be that PhDs in computer bly use them less, and he would rather science here get rubber chickens see them in constant use. At any rate, thrown at them at the reception in I sent him some email congratulating their honor immediately after their him on the 25th birthday of his inven- thesis defense. Some time later I men- tion; in his response he said that the tioned that Mary Harper, one of my stu- CMU publicity people had had the idea dents, is now giving rubber chickens to

conduit! 22 her PhD students. A few weeks ago I learned that an- other of my students, Sharon Caraballo, has also been en- gaged in some fowl play (I’m sorry, I am not strong enough to resist). Sharon took her chicken to a taxidermist and had it stuffed and mounted! It is now proudly displayed on a wooden plaque above her desk at her office at Geor- getown. She explained that the wooden plaque is actual- ly one for fish, since the rub- ber chicken is too small for anything else. She also told Having just defended his thesis and received his PhD me that when she took it to poultry, Air Force Major Joel Young is a happy man! the taxidermist she gave it to “:-),” would not be the correct smiley. him with a straight face, no explana- That is, I should not have written ““:-)”,,” tion. I can only try to imagine what the but rather ““:-),”.” (Everyone got that?) guy was thinking. However, she was This convention, even when applied to accompanied by her husband when normal text, hardly ever makes sense. she went to pick it up, and her hus- Consider the sentence: I said “maybe.” band took pity on the fellow and told As I have noted in previous conduit!s, him what was going on. The taxider- I have been doing research on getting mist said he had probably gotten more computers to establish the correct sen- comments on it than anything else he tence structure for sentences such as had ever done. that. This involves creating tree-struc- In case you want to have your chicken tures that explicitly indicate structure in mounted, you might want to take it to terms of noun-phrases (NPs), verb someone with experience, so Sharon phrases (VPs), etc. Now if English were has given me the fellow’s address: logical, we could have written the above Taxidermy Unlimited sentence as: I said “maybe”. Vienna, VA Then we could nicely write out the 703-255-1460 tree-structure as follows: Tell him “conduit! sent me.” Those of you not interested in details S of the English language can skip the rest of this article, as I feel the need to fess up to some ungrammaticalities in NP the first three paragraphs of this edi- VP tion of “unplugged.” As is well known to all copy editors but only moderately known to the rest of us, commas and fi- I said NP . nal punctuation go inside final quota- tion marks, not after, as placed in the “ maybe” previous sentence. This can be a real bother. So in the first paragraph I had a situation where my smiley “:-)” came Note how the quotation marks go with the just before where a comma should oc- noun phrase (they help describe what was cur. Thus I should have put the comma said) while the final punctuation goes with inside the quotes, but I did not, because the complete sentence (S), as it should.

conduit! 23 However, with the nonlogical version we are but I plead with you to see my point of view! To left with two very unpalatable alternatives: get my parser to work I use a large corpus S S (one million words) of hand-annotated syn- tactic structures. This VP corpus, for lack of any- NP VP NP thing better, uses the first of the above ter- rible messes. Howev- NP . NP er, I am now trying to I said ” I said apply this work in machine translation, “ maybe “ maybe . ” and since other lan- guages do not share English’s mess, we en- In one case the NP ends up having only a single counter a problem when trying to figure quotation mark and the S gets the other, while in out how quotations should be translated. the second case the NP gets both quotation marks This causes no end of grief. The English but also the period, which makes no sense. convention has to be reformed. And, you Now this may strike you as completely unimportant, can quote me, “immediately!”

conduit! A publication of The Computer Science Department Brown Uiversity

Inquiries to: conduit! Department of Computer Science Box 1910, Brown University Providence, RI 02912 Fax:.401-863-7657 Phone:.401-863-7610 Suzi Howe Katrina Avery Jeff Coady email:[email protected] Editor-in-Chief Editor Technical Support WWW:.http://www.cs.brown.edu/ 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