2000-2001 AbstractsAWICS Series:sponsored by MicrosoftBringing Order to the Web and BeyondSusan T. Dumais,Microsoft Research 4:10pm, Monday October 2, 2000
AbstractThis talk describes algorithmic and interface innovations to help users organize information, in particular web search results. Today web search services returned a ranked list of best-matching pages and users have to sift through long undifferentiated lists. An important mechanism for facilitating information access in a wide variety of applications is a structure knowledge hierarchy, such as those used by library classification systems and more recently web directories like Yahoo! and LookSmart. In this talk I will describe how such structure can be used to automatically organize web search results. For example, a query about "saturn" will group the returned pages into those having to do with automobiles, computer games, and outer space. There are two key technologies for doing this: 1) developing models for hierarchically classifying arbitrary text pages on-the-fly, and 2) building an interface for taking advantage of the resulting structure. I will begin by talking about enhancements to support vector machine (SVM) learning algorithms for text classification that exploit hierarchical structure. I will then describe how we developed and evaluated a series of user interfaces to support structured search. Category interfaces that show results in context (not just results plus context) provide retrieval advantages of up to 50% and strong user preferences.
BiographySusan Dumais is a Senior Researcher in the Adaptive Systems and Interaction Group at Microsoft Research where she works on algorithms and interfaces for improved information access and management. Prior to joining Microsoft Research in July 1997, she was at Bellcore and Bell Labs for seventeen years where she worked on a wide variety of human-computer interaction and information retrieval projects. She received a B.A. in Mathematics and Psychology from Bates College and a PhD in Cognitive Psychology from Indiana University.She has published widely in the areas of human-computer interaction and information retrieval. Her recent research activities have focused on hierarchical text categorization using inductive learning approaches, and collaborative information retrieval (a joint project with Univ Washington, Boeing and RISO Labs). Previous research included well-known work on a statistical method for concept-based retrieval known as Latent Semantic Indexing, interfaces for combining search and navigation, user modeling, individual differences, and organizational impacts of new technology. Susan is Chair of ACM's SIGIR group, and serves on the NRC Committee on Computing and Communications Research to Enable Better Use of Information Technology in "Digital Government". She serves on the editorial board of: Information Retrieval, ACM:Transactions on Information Systems, Human Computer Interaction, Information Processing and Management, Hypertext, Encyclopedia of Information Retrieval, and Annual Review of Information Science and Technology, and is actively incolved on program committees for several conferences. She has been a visiting faculty member at the University of Chicago, New York University, and Stevens Institute of Technology.
Faculty Contact:Nancy Amato (amato@cs.tamu.edu)
AWICS Series:sponsored by CRA-W and Lucent TechnologiesAnimating with SimulationJessica Hodgins,Carnegie Mellon University 4:10pm, Wednesday November 15, 2000
AbstractComputer animations and virtual environments both require a source of motion for characters and objects in the environment. We are exploring simulation as a possible solution to this problem. For characters, this solution requires applying control algorithms to physically realistic models of the systems that we would like to animate. For objects in the environment, this solution requires modeling the physics of the situation. By using these techniques to simulate humans, we are working towards avatars that are responsive to the user's subtle gestures and interactive agents that respond appropriately to events in a virtual environment. For example, we have developed control algorithms that allow rigid body models to run or bicycle at a variety of speeds, bounce on a trampoline, and to perform a handspring vault and several platform dives. Recently, we have begun to use human data to inform the behavior of the control systems. We have also modeled complex physical properties of materials such as fracture and explosions. Because our goal is natural looking motion, we compare the computed motion for each simulation to that of humans or objects in the real world.
BiographyJessica Hodgins joined the Robotics Institute and Computer Science Department at Carnegie Mellon University as a Associate Professor in fall of 200. Prior to moving to CMU, she was an an Associate Professor and Assistant Dean in the College of Computing at Georgia Institute of Technology. She received her Ph.D. in Computer Science from Carnegie Mellon University in 1989. Her graduate research involved programming a two-legged laboratory robot to run, accurately place its foot to avoid obstacles, climb stairs, and perform gymnastic maneuvers. Her current research focuses on computer graphics, animation, and interactive virtual environments. Her research explores techniques that may someday allow robots and animated creatures to plan and control their actions in complex and unpredictable environments. She has received a NSF Young Investigator Award, a Packard Fellowship, and a Sloan Fellowship. She is editor-in-chief of ACM Transactions on Graphics.Faculty Contact: Nancy Amato (amato@cs.tamu.edu)
Shell Series:
Compiler Technology and Scientific Computing
David Padua, 4:10pm, Monday February 19, 2001
AbstractScientific computing has always been an important focus for compiler technology, to the point that the first commercial compiler was a Fortran compiler and many of the early optimization techniques were developed to improve the performance loops containing array accesses. Today, powerful compilers are an important part of any development environment for numerical codes. These compilers usually generate very good code that can take advantage of the capabilities of the most powerful processors. Although much has been achieved since the first Fortran compiler, there are still many interesting and important issues in compiler technology that need attention. For example, techniques to detect coarse grain parallelism are quite important and more needs to be done in this area, which has proven much more difficult than originally expected. It is also important to systematize the optimization process, which traditionally has been based on ad-hoc procedures that are not well understood, and the development of compiler algorithms that take advantage of information available in very high level programs to perform advanced optimizations. As an illustration of research impacting the last two issues, I will discuss two compilers for linear algebra languages under development at Illinois. The first is a MATLAB compiler. MATLAB codes can be very inefficient when interpreted; but, when translated by an effective compiler, their performance usually matches that of equivalent Fortran codes. A batch MATLAB-to-Fortran 90 system was completed at Illinois a few years ago and we are now working on a Just-In-Time extension that we call MAJIC (Matlab Just-In-time Compiler). The techniques and performance results of both strategies will be presented. The second compiler to be discussed accepts tensor product formulations of signal processing algorithms and generates programs that implement them. By applying mathematical identities, a given formula can be transformed into a large number of equivalent forms. The tensor product compiler (TPC) searches the space of possibilities for the formula and program forms that give the best performance. The translation and search techniques used by TPC and their possible applications in more general contexts will be discussed.
BiographyDavid A. Padua received the Ph.D. degree from the University of Illinois at Urbana-Champaign in 1980. In 1985, he returned to the University of Illinois where, from 1990 to 1993, he was Associate Director for Software of the Center for Supercomputing Research and Development and is now a Professor in the Department of Computer Science. Dr. Padua's interests are in parallel computing, including machine organization, programming languages and tools, and compilers. His current research focuses on the experimental analysis of compilers and on the development of techniques to make compilers more effective.Faculty Contact: Lawrence Rauchwerger (rwerger@cs.tamu.edu)
Shell Series:
Developing and Integrating Software/System Product, Process, Property, and Success Models
Barry Boehm, 10:00am, Thursday March 22, 2001
AbstractAt the USC Center for Software Engineering, we have been doing research on software/system process models (WinWin Spiral Model, Life Cycle Anchor Points; Schedule as Independent Variable); product models (domain models, requirements, architectures); property models (COCOMO II and its COTS integration, quality, and Rapid Application Development extensions); and success models (stakeholder win-win, business-case, IKIWISI -- I'll know it when I see it -- prototyping). This talk will begin by summarizing highlights of our research results, their applications, and synergies among the classes of models.In exploring the interactions among the classes of models, we have found that a major explanation for the "tar-pit" nature of large software projects is that they unwittingly inherit "model clashes" among their product, process, property, and success models. Some frequent examples are clashes between requirements-driven processes and COTS-driven products; or using a develop-to-cost process with unprioritized requirements or a tightly-coupled architecture, both of which make it hard to quickly drop features when you start running out of budget. We have been developing and testing an approach called Model-Based (System) Architecting and Software Engineering (MBASE), with some pretensions toward being a "unified field theory" of how to reconcile software and system product, process, property, and success models. The talk will also summarize the MBASE approach and our experiences in applying it on over 100 digital library projects, and on its emerging use in government and industry.
BiographyBarry W. Boehm, TRW Professor of Software Engineering and Director, Center for Software Engineering, University of Southern California Barry Boehm received his B.A. degree from Harvard in 1957, and his M.S. and Ph.D. degrees from UCLA in 1961 and 1964, all in Mathematics. He also received an honorary Sc.D. from the U. of Massachusetts in 2000. Between 1989 and 1992, he served within the U.S. Department of Defense (DoD) as Director of the DARPA Information Science and Technology Office, and as Director of the DDR&E Software and Computer Technology Office. He worked at TRW from 1973 to 1989, culminating as Chief Scientist of the Defense Systems Group, and at the Rand Corporation from 1959 to 1973, culminating as Head of the Information Sciences Department. He was a Programmer-Analyst at General Dynamics between 1955 and 1959.He has been a member of NASA's Research and Technology Advisory Committee and Chair of its Advisory Committee on Guidance, Control, and Information Systems; and a member of the Air Force Scientific Advisory Board and Chair of its Information Technology Panel. He currently serves as Chair of the Board of Visitors for the CMU Software Engineering Institute. He has served on the boards of several companies and scientific journals, and as a member of the Governing Board of the IEEE Computer Society. He is an AIAA Fellow, an ACM Fellow, an IEEE Fellow, an INCOSE Fellow, and a member of the National Academy of Engineering. Faculty Contact: Hoh In (hohin@cs.tamu.edu)
Shell Series:
Visualization of Multiple-criterion Decision Problems and
Search for Efficient Decisions
Alexander Lotov, 4:10pm, Wednesday April 11, 2001
AbstractThe graphic multiple-criterion method, the Feasible Goals Method is described. It supports a search for preferable decisions in the case of conflicting criteria. The method transforms a mathematical model of a decision problem into colour pictures, which inform users on the variety of feasible criterion vectors and on efficient tradeoffs. The pictures are displayed on-line and are animated. Due to this, efficient tradeoffs for up to seven criteria can be assessed by user. This knowledge helps to identify the preferable criterion vector (the feasible goal vector) directly on a picture and to receive the related decision. So, the FGM differs from the well-known methods based on multi-attribute utility theory as well as from outranking methods, interactive optimization-based procedures, etc. Concept of the FGM, its mathematical basis and the experience of its real-life application in Russia for development of water quality improvement projects are considered. Its possible application in the framework of asynchronous network preparation of decision meetings is discussed. Further development of the FGM, the Reasonable Goals Method (RGM) is based on the visualization large databases. The data are enveloped, and a goal is identified among the envelope points (reasonable goal). Applications of the RGM and the RGM-based Internet application server are outlined.
BiographyProfessor Alexander V. Lotov has been head of the Computing Center of the Russian Academy of Sciences since 1990. He was a researcher at the International Institute for Applies Systems Analysis in Austria in 1981, and was named a Leading Research Fellow in 1988-1990. He is a member of the Council and of the Governing Body of the Russian Operations Research Society. He has authored or coauthored four books on mathematical modeling of economic systems and decision making, and has published research papers in many leading journals. In 1997 he received the Medal in Commemoration of 850 Years of Moscow City, and in 2000 he received the Edgeworth-Pareto Award of the International Society for Multiple Criteria Decision Making. He has lectured extensively in the US and in Europe.Professor Lotov's research applies operations research/management science, human/computer interaction, graphic decision and negotiation support, and other mathematical and computer tools to the assessment of environmental problems and public decision making. His work has included the use of the internet for public decision making about environmental and economic problems, visual data analysis and decision analysis applied to computer networks, mathematical problems of decision support, approximation and perturbations of multiple dimensional sets, linear inequalities, and dynamic systems. Faculty Contact: Hoh In (hohin@cs.tamu.edu)
AWICS Series:sponsored by MotorolaProphesy: An Infrastructure for Analyzing and Modeling the Performance of Parallel and Distributed ApplicationsValerie Taylor,Northwestern University 4:10pm, Monday April 30, 2001
AbstractEfficient execution of applications requires insights into how system features impact the performance of the application. The availability of national, high-speed networks has made available distributed systems for execution of large-scale applications. Distributed systems, however, consists of heterogeneous components, such as networks, processors, run-time systems, operating systems, etc. This heterogeneity complicates the task of gaining insights into the performance of the application.This talk presents the Prophesy Project, an infrastructure that aids in gaining this needed insight based upon one's experience and that of others. Prophesy consists of three major components: a relational database that allows for the recording of performance data, system features and application details; an application analysis component that automatically instruments applications and generates control flow information; and a data analysis component that facilitates the development of performance models, predictions and trends. As a result, the Prophesy system can be used to develop models based upon significant data, identify the most efficient implementation of a given function based upon the given system configuration, explore the various trends implicated by the significant data, and predict the performance on a different system.
BiographyValerie E. Taylor received her B.S. in Computer and Electrical Engineering and M.S. in Electrical Engineering from Purdue University in 1985 and 1986, respectively. She received her PhD in Electrical Engineering from University of California at Berkeley in 1991.Valerie E. Taylor is an Associate Professor in the Electrical and Computer Engineering Department at Northwestern University and holds a guest appointment with the Mathematics and Computer Science Division at Argonne National Laboratory. She has been at Northwestern since 1991. Her research interests are in the areas of computer architecture and high performance computing, with particular emphasis on mesh partitioning for distributed systems and the performance of parallel and distributed applications. In 1993, Valerie Taylor received a National Science Foundation ``National Young Investigator'' award. She holds a U.S. patent for her dissertation work on sparse matrices. She also has a copyright for the RAB tool developed at Purdue University. She is a member of the Association for Computing Machinery, Society of Industrial and Applied Mathematics, and the Institute for Electrical and Electronics Engineers. She is a member of the SC Steering Committee and Co-Chair of the Coalition to Diversify Computing. Faculty Contact: Nancy Amato (amato@cs.tamu.edu) |
| ||||
| ||||