Cooperative Research
Cooperative Research |
|
 |
 |
 |
 |
 |
zack@dri.cornell.edu http://dri.cornell.edu
The Design Research Institute (DRI), a partnership of Xerox Corporation and Cornell University, brings computer and information science and high-performance computing technology to bear on problems of engineering design that are important to American industry. Research in DRI focuses on technologies for shortening the product development cycle, particularly for complex electro-opto-mechanical systems. In DRI, Xerox scientists on campus collaborate with Cornell faculty and students in three broad areas: information capture and access, collaboration technology, and computational simulation of physical systems. Most of the research projects are not proprietary and are candidates for external funding and for participation by other companies. Research in information capture and access involves pathways and repositories for engineering information, documents, and knowledge. An important element of this work is Dienst, a software system for organizing and providing World-Wide-Web access to document collections anywhere on the Internet. Dienst, being co-developed by Xerox and Cornell, has attained worldwide recognition and widespread use as a major platform for digital library development. DRI research in collaboration technology centers on LARC, a joint project with Digital Equipment Corporation. LARC is developing information capture and sharing technologies to support engineering while simultaneously optimizing group work processes and laboratory and office work settings. In conjunction with an advanced technology group in the Xerox Wilson Center for Research and Technology in Webster, NY, LARC has created an innovative work environment featuring World-Wide-Web-based information sharing, innovative use of video to support lab experiments, and a project center based on Xerox Liveboards that serves as a hub of collaborative activity. A similar system has been deployed in the Design Studio of the Future in the Cornell Center for Manufacturing Enterprise, making use of the same technology in an educational setting. The study of the design process itself is an important area of collaboration technology in which computation merges with document formalism. A concept of "active design documents" is being developed for creating software agents embedded in documents. The agents compute design tradeoffs to assist engineers in the design process. With DARPA support, DRI is developing software that provides interoperability of heterogeneous databases to support concurrent engineering. A central part of this work is the development of a Metadata Dictionary, a repository of schema descriptions used by automatic data translators. With the Cornell Theory Center, DRI is exploring access to very large scientific and engineering databases. Computational simulation research in DRI is focused on silicon micro-electro-mechanical structures such as those used in thermal ink-jet printing nozzles or in micro-actuators and sensors. Analyzing such systems by computer instead of by building hardware prototypes greatly reduces design time. Of special interest to DRI is developing World-Wide-Web access to simulations executed on parallel supercomputers in the Theory Center. With industry scientists engaged in collaborative research on campus, the DRI brings an awareness of the needs of industry to Cornell, facilitates the transfer and use of research results, and enhances the access of new graduates to career opportunities in industry.
coleman@cs.cornell.edu http://www.cs.cornell.edu/Info/Annual96/ACRI.html
The Advanced Computing Research Institute (ACRI), under the direction of Thomas F. Coleman, is concerned with scientific computation research and its application to engineering and scientific problems. Of particular importance is the use of advanced computer architectures and environments. The ACRI, a unit of the Cornell Theory Center, is closely connected with the Computer Science Department through several faculty members and researchers.
Current research projects include: the design and application of efficient numerical algorithms for continuous optimization problems, parallelizing compilers for scientific computation (the Bernoulli project), the design of parallel algorithms for linear algebra and signal processing, the understanding and development of new methods for the numerical solution of differential equations, automatic grid generation, large-scale computational differentiation, and large-scale inverse problems (e.g. image enhancement, tomography).
A major research emphasis of the ACRI over the past year has been the Bernoulli project. The Bernoulli project is implementing a restructuring compiler that generates efficient parallel code for sparse matrix problems, starting from high-level specifications of the dataflow of the underlying algorithms. The first milestone is the generation of parallel code for preconditioned iterative methods for solving large sparse systems of linear equations. After that, we will tackle direct methods incorporating multifrontal technology. This compiler is targeted toward applications in which static decomposition of data is appropriate. For applications like adaptive mesh refinement (AMR) codes in which computation and communication requirements cannot be determined statically, we are implementing a runtime system based on efficient communication and load balancing. Some of the code for the runtime substrate is being distributed widely to the HPC community through our participation in the PORTS consortium.
Research activity has also centered around the development of a very high-level special-purpose "language'' for computations involving Fast Fourier Transformations (FFTs). This language, based on the mathematical notion of a Kronecker product, allows for short simple expression of FFT-related programs arising in signal processing. Efficient extraction of parallelism is one of the many benefits of this approach.
In addition to basic algorithmic and systems research, work continues on several collaborative/application fronts. For example, a joint effort on large-scale nonlinear structural mechanics, with Professor Tim Healey in Cornell's Theoretical and Applied Mechanics (TAM) Department, was initiated this year. Chris Wohlever, a graduate of TAM, joined ACRI to investigate the parallel computation of problems in structural mechanics-in particular, the "buckling" problem. Wohlever's work involves the use of Multi-Matlab, a Computer Science/Theory Center research development. Continuing collaborative/application work includes the application of inverse techniques and optimization to ultra-sound imaging problems, global (parallel) minimization methods for molecular conformation problems, and the application of the techniques of automatic differentiation to discrete optimal control problems.
Work on unstructured mesh generation continues. The QMG 1.0 software package, developed in the ACRI and released on the Web in May 1995, is currently in use at dozens of sites in the U.S. and Europe. This package automatically generates 3D volumetric finite element meshes for polyhedral objects with complicated topology, including internal boundaries. QMG is slated for inclusion in an upcoming release of Ellpack, a finite element package from Purdue. Work continues on generalizing QMG and improving the basic algorithms.
dean@cs.cornell.edu http://www.cs.cornell.edu/home/dean/
The Cornell Digital Library Research Group (CDLRG) is dedicated to creating open, interoperable standards for providing digital library services over the global Internet. The CDLRG is an outgrowth of the DARPA-funded Computer Science Technical Reports Project (CS-TR) project. The CS-TR project is a collaboration of the five leading U.S. computer science departments - Berkeley, Carnegie-Mellon, Cornell, MIT, and Stanford - and the Corporation for National Research Initiatives (CNRI), with the goal of developing methods for creating a networked digital library of computer science technical reports and similar "gray" publications.
As part of the CS-TR project, researchers in the CDLRG and the Xerox Design Research Institute (DRI) developed Dienst, an architecture and protocol for distributed digital document libraries. The notable features of Dienst are:
* An open protocol that supports a number of distributed library services-repositories for document storage, indices for document search, and user interfaces for a human front-end to these services.
* A document model in which documents have globally unique names, have multiple formats (e.g. PostScript, archival TIFF, OCR output), and can be retrieved in either physical (pages) or logical (e.g. chapters, sections) decompositions.
* A distributed search architecture in which multiple distributed search engines are searched in parallel and results are combined at the user interface layer into a unified "hit" list.
An early implementation of Dienst was installed at all the participating CS-TR computer science departments, providing uniform networked access to their combined technical reports collections. As part of this effort, the Cornell Computer Science Department scanned its entire technical reports collection, from 1968 through the present, and made them available through the Dienst server.
The CS-TR project has evolved into the Networked Computer Science Technical Reports Library (NCSTRL) project, for which the CDLRG is the central research, development, and support site. NCSTRL is an international consortium with three primary goals:
1. It is a production computer science technical reports library. Using an extended version of the Dienst architecture, nearly sixty computer science departments and research centers worldwide cooperate in the NCSTRL collection. NCSTRL is used by thousands of researchers from around the world to search for, browse, read, and download the technical reports from these institutions.
2. NCSTRL is a foundation for the testing and demonstration of current digital library research. The CDLRG is cooperating with a number of other departments and research centers who wish to use the NCSTRL collection in this manner.
3. NCSTRL is a laboratory for exploration of and experimentation with many of the non-technical issues concerning distributed digital libraries. These include intellectual property issues and concerns about collection integrity and service integrity.
In the context of NCSTRL and in collaboration with CNRI and the joint NASA/ARPA/NSF Digital Library Initiative sites, the CDLRG is currently undertaking research in a number of fundamental digital library issues. This research focuses on four areas.
* What is the service structure of a distributed digital library and what is the interaction between these services? In particular, CDLRG researchers are concerned with the structure of and interface to a repository, which allows for deposit of and access to objects in digital form (digital objects). Repositories are administrative entities that play a primary role in protecting the intellectual content of the contained objects.
* How are intellectual property rights protected in a digital library environment? In particular, what is the language for expressing the terms and conditions for access to digital objects and how are these terms and conditions associated with the objects?
* How can the digital library architecture accommodate the diverse forms of descriptive metadata that should be associated with objects? This metadata includes cataloging forms such as MARC, domain-specific forms such as CSDGM for geospatial data, and other forms expressing information about provenance, linkage, administration, structure, and the like. Not only are these different metadata forms logically distinct, but they are created and administered by separate authorities, subject to different access control rules, and stored in diverse forms including ASCII, binary, and even executable objects such as Java applets.
* What are the methods for efficient and reliable distributed search? Whereas centralized indexing is far simpler, it will neither scale nor fit comfortably into a world where information may be partitioned among several indexing sites due to ownership or separate administration. Moreover, we can expect for the foreseeable future that the global Internet will have unpredictable and unreliable performance characteristics. In light of this, we must develop methods for mirroring, caching, and partitioning information so that response time for search and retrieval is reasonable.
As this research progresses, results will be implemented as future extensions to the Dienst architecture and deployed in the NCSTRL framework.
The CDLRG consists of three full-time staff: Carl Lagoze, the Project Leader, who provides day-to-day direction for research and operations; David Fielding, Senior Systems Programmer; and Meredith Curtin, User Support Specialist. Overall leadership for the project is provided by Dr. Dean B. Krafft. The CDLRG has also benefited greatly from the contributions of James R. Davis, of the Xerox Design Research Institute.
The Computer Science Department plays a very visible role in the university's Cognitive Studies Program. The year-long Cognitive Studies Proseminar (COGST 773/4; CS 773/4) traditionally begins with an introduction to the computational aspects of cognitive science: in the fall of 1995, six CS faculty and researchers ( Juris Hartmanis, David Gries, Stuart Allen, Ramin Zabih, Amit Singhal, and Claire Cardie) provided weekly lectures to introduce the theoretical and methodological issues that underlie computer science and that link Computer Science with Linguistics, Philosophy, and Psychology. In addition, Ramin Zabih and Claire Cardie helped to direct the undergraduate Cognitive Studies Concentration. Together with Barbara Lust (Linguistics and HDFS), they also run the NSF-sponsored Graduate Research Training Program in Cognitive Science. Each year, this program supports approximately 5 Ph.D. students who plan to pursue interdisciplinary research problems that connect Computer Science with other disciplines in Cognitive Studies. In addition to faculty support of the Cognitive Studies Program, graduate student interest in the field continued to grow-seven of twenty graduate students in the 1995-1996 Proseminar were from Computer Science. In addition, three CS students ( Rick Aaron, Julia Komissarchik, and Scott Mardis) had abstracts selected for presentation at the Annual Cognitive Studies Graduate Student Research Forum held on October 21st at the Miller-Heller House.
The Computer Science and Electrical Engineering departments have significantly increased their collaborations and joint planning activities in recent years. These research and curriculum development activities are focused primarily in the broad area of information technology. Recent hires in both units - Zygmunt Haas, Sheila Hemami, Brian Smith, Thorsten von Eicken, and Ramin Zabih - are building joint research programs among themselves, as well as expanding existing research projects. Efforts include establishing a joint Visual Communications lab and installing a video infrastructure to link remote offices and labs. These efforts are not limited to research programs. Curriculum development initiatives include planning a joint undergraduate computing laboratory and an Information Technology Laboratory. Professors Hemami and Zabih are developing a course on Image Processing and Computer Vision. New hires Srinivasan Keshav, Praveen Seshadri and Stephen Wicker create additional bridges between EE and CS research and industrial programs.
Go to:
Research Interests of
Faculty and Senior Researchers
Return to:
1996-1996 Annual Report
Home Page
Departmental Home Page
If you have questions or comments please contact: www@cs.cornell.edu.