 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
 |
|
||||||||||||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
 |
 |
 |
 |
|
|||||||||||
 |
 |
 |
|
||||||||||||
|
|
|
|
|
 |
|
|||
 |
 |
|
||
 |
|
|||
 |
|
|||
 |
|
|||
 |
|
|||
 |
|
|||
 |
|
|||
 |
|
|||
 |
|
|||
 |
 |
 |
|
|
 |
|
|||
The 2000-2001 Jon Postel
Distinguished Lectures
JITENDRA MALIKUC Berkeley DATE: Thursday, November 9,
2000
ABSTRACTVision has been defined as the process of knowing what is where by looking. The geometrical aspects of computer vision, the “Where” question, have seen considerable progress and problems such as inferring three-dimensional structure from multiple views are now well understood. The “What” question has proved much more elusive—recognizing objects (This is my poodle, Fido) or categories (This is a dog) or activities/expressions (He is jumping) – are hard problems where machine capabilities fall far short of humans. The challenge comes from the fact that recognition has to be performed in spite of considerable variations in viewpoint, lighting and from one single objects in isolation and part of the problem is determining what set of pixels belong to a single object. We approach this as a two-stage process: a process of image segmentation – grouping pixels to form regions of coherent color and texture, and a process of recognition – comparing assemblies of such regions, hypothesized to correspond to a single object, with views of stored prototypes. We treat segmenting images into regions as an optimization problem: partition the image into regions such that there is high similarity within a region and low similarity across regions. This is formalized as the minimization of the normalized cut between regions. Using ideas from spectral graph theory, the minimization can be set up as an eigenvalue problem. The resulting eigenvectors provide a hierarchical partitioning of the image into regions. Visual attributes such as color, texture, contour and motion are encoded in this framework by suitable specification of graph edge weights. The recognition problem requires us to compare assemblies of image regions with previously stored prototypical views of known objects. The comparison is based on shape, color and texture. Of these the most challenging problem is that of shape matching. We have devised a novel algorithm for shape matching based on a relational descriptor called the shape context. This enables us to compute similarity measures between shapes which, together with similarity measures for texture and color, can be used for object recognition. The shape matching algorithm has yielded excellent results on various tasks such as handwritten digit recognition and trademark similarity searches. This talk draws on joint work with Serge Belongie, Thomas Leung, Jan Puzicha and Jianbo Shi.
FRANS KAASHOEKMIT DATE: Tuesday, January 23,
2001
ABSTRACTMany software systems are inflexible, limiting the performance and functionality of applications running on top of them. These inflexible systems offer only large high-level abstractions, making it difficult or impossible to provide different semantics or different implementations. In several systems we have built at MIT, we have achieved flexibility by adhering to a strong variant of the end-to-end argument (Saltzer et al. 1984). In this talk, I will discuss our interpretation of the end-to-end argument and how we applied it to the design of a flexible operating system, a secure file system, and a modular router. BIOGRAPHYM. Frans Kaashoek is an associate professor in MIT’s EECS department and a member of the Laboratory for Computer Science. He received a PhD from the Vrije Universiteit (Amsterdam, The Netherlands) for his work on group communication in the Amoeba distributed operating system His principal field of interest is designing and building computer systems.
JAMES E. SMITHUniversity of Wisconsin Madison DATE: Thursday, February 22,
2001
ABSTRACTWithin two or three technology generations, processor architects will face a number of major challenges. Wire delays will become critical, and power considerations will temper the availability of billions of transistors. Many important software applications will be object-oriented, multithreaded, and will consist of many separately compiled and dynamically linked parts. To accommodate these shifts in both technology and applications, microarchitectures will process instruction streams in a distributed fashion - instruction level distributed processing (ILDP). ILDP will be implemented in a variety of ways, including both homogeneous and heterogeneous elements. To help find run-time parallelism, orchestrate distributed hardware resources, implement power conservation strategies, and to provide fault-tolerant features, an additional layer of abstraction - the virtual machine layer - will likely become an essential ingredient. By providing the architect with a layer of software, a cohesive virtual machine can be designed to provide close hardware/software interaction and runtime optimization.
V.S. SUBRAHMANIAN* Norman E. Friedmann Distinguished Lecture * University of Maryland, College
Park DATE: Thursday, March 8,
2001 IMPACT (Interactive Maryland Platform for Agents Collaborating Together) provides a platform for the creation and deployment of distributed, multiagent applications by building on top of legacy and/or specialized codebases. In this talk, I will describe the overall architecture of the IMPACT system, and outline how this architecture:
I will use a supply chain automation example to illustrate the concepts and (if possible) will provide a demonstration. The specific work described in the talk is joint work with Thomas Eiter.
Return to: Index / 1995-1996 Series / 1996-1997 Series / 1997-1998 Series / 1998-1999 Series For further information, call the UCLA Computer Department at (310) 825-3886. |
|||||||||||||||||||||||||||||||||||||||||||
|
