Return to BSD News archive
Newsgroups: comp.os.386bsd.development
Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.Hawaii.Edu!ames!olivea!decwrl!decwrl!csus.edu!netcom.com!jmonroy
From: jmonroy@netcom.com (Jesus Monroy Jr)
Subject: Seminar on Computing
Message-ID: <jmonroyC8pM67.3MH@netcom.com>
Keywords: Computer Operating system VLSI reliability
Organization: NETCOM On-line Communication Services (408 241-9760 guest)
Date: Wed, 16 Jun 1993 10:22:55 GMT
Lines: 70
I thought this might be of interest to everyone.
The Seminar seems to be given twice in one day.
The listed is the morning program (I assume)
and what follows seems to be for the afternoon.
UC Santa Cruz: Computer & Information Sciences, and
Computer Engineering
RESEARCH SEMINAR: Tuesday, June 22, 1993, 2:00 - 4:00 PM
Room 330, Applied Sciences Building
==========================================================================
From: gerri@cse.ucsc.edu (Gerri McLellan)
Newsgroups: ba.seminars,ucsc.baskin.general
Subject: UCSC CE/CIS Research Seminar 6/22/93
Date: 14 Jun 1993 15:22:43 GMT
Organization: University of California, Santa Cruz (CE/CIS Boards)
UC Santa Cruz: Computer & Information Sciences, and
Computer Engineering
RESEARCH SEMINAR: Tuesday, June 22, 1993, 10:30 - 11:30 AM
Room 330, Applied Sciences Building
Prof. Andrew Kahng, Computer Science Department, University of
California, Los Angeles, will speak on ``New Directions in Practical
Large-Scale Optimization''
Abstract:
Large instances of intractable optimizations such as scheduling,
partitioning or quadratic assignment confront us in every phase of
VLSI system synthesis. However, current optimization methods cannot
guarantee any prescribed level of solution quality, and are not easily
tuned to prescribed time limits for the optimization.
This talk surveys three areas of recent work which shed new light on
these issues.
First, we have studied new models for how the optimization process
actually takes place. Our "best-so-far" paradigm has significant implications
for the "simulated annealing" (SA) heuristic, which has been applied to
every single hard optimization in VLSI CAD. All existing SA implementations
use monotone "cooling" temperature schedules which are motivated by the
annealing algorithm's proof of optimality as well as by analogies with
statistical thermodynamics. However, best-so-far analysis leads to
revisionist intuitions vis-a-vis the justification and implementation of
simulated annealing, and moreover leads to improved solution qualities
in practice. Second, we have continued to develop and exploit
structural models for large real-world optimizations; our most recent results
include new non-monotone hill-climbing approaches which have led to effective
time-bounded optimization. Finally, we have developed insights into
the relationship between problem structure and amenability to traditional
optimization approaches such as hill-climbing or greedy multi-start
(e.g., the Kernighan-Lin algorithm). For certain problem classes (traveling
salesman and graph bisection), we achieve dramatic CPU savings with no
loss of solution quality.
------------------------
Anyone needing special arrangements to accommodate a disability is
encouaged to call Gerri McLellan at the Baskin Center (408) 459-3695.
___________________________________________________________________________
Jesus Monroy Jr jmonroy@netcom.com
/386BSD/device-drivers /fd /qic /clock /documentation
___________________________________________________________________________