clarke@utcsri.UUCP (10/09/87)
(SF = Sandford Fleming Building, 10 King's College Road)
(GB = Galbraith Building, 35 St. George Street)
(Sorry this is late!)
SUMMARY:
A.I. SEMINAR, Tuesday, October 13, 2 pm, GB244 -- Phil Agre:
"Pengi: An Implementation of a Theory of Activity"
GRAPHICS SEMINAR, Tuesday, October 13, 3 pm, GB120 -- Jane Wilhelms:
"Dynamics for Computer Animation"
SYSTEMS SEMINAR, Thursday, October 15, 11 am, SF1101 -- Oscar M. Nierstrasz:
"Active Objects"
NUMERICAL ANALYSIS SEMINAR, Thursday, October 15, 3:30 pm, GB119 -- W.P. Tang:
"Schwarz Splitting and Template Operators"
SYSTEMS SEMINAR, Friday, October 16, 11 am SF1105 -- Satish K. Tripathi:
"Resource Allocation in Distributed Systems"
THEORY SEMINAR, Friday, October 16, 2 pm, GB120 -- Mauricio Karchmer:
"A Monotone Depth Lower Bound For Transitive Closure"
----------------------------------------
A.I. SEMINAR, Tuesday, October 13, 2 pm, GB244
Mr. Phil Agre
MIT AI LAB
"Pengi: An Implementation of a Theory of Activity"
AI has generally interpreted the organized nature of everyday activity
in terms of plan-following. Nobody could doubt that people often make and
follow plans. But the complexity, uncertainty, and immediacy of the real
world require a central role for moment-to-moment improvisation. Before
and beneath any planning ahead, one continually decides what to do now.
Investigation of the dynamics of everyday routine activity reveals impor-
tant regularities in the interaction of very simple machinery with its
environment. We have used our dynamic theories to design a program, called
Pengi, that plays a video game called Pengo in which a player controls a
penguin in a battle against deadly sno-bees. Pengi engages in complex,
apparently planful activity without requiring explicit models of its world.
My presentation will focus on Pengi's novel account of representation.
Pengi can use generalized knowledge without using variables, so its
machinery can be much simpler than that of production systems or first-
order theorem-provers. Specifically, Pengi is composed of a periphery
(visual and motor systems) much as described by Marr and Ullman and a
center made entirely of combinational logic. Pengi's center is, in partic-
ular, a connection network, and indexical-functional representation offers
a solution to (at least some cases of) the connectionists' binding problem.
GRAPHICS SEMINAR, Tuesday, October 13, 3 pm, GB120
Dr. Jane Wilhelms
University of California
"Dynamics for Computer Animation"
A major problem in computer animation is creating motion that appears
natural and realistic. Most animation at present is produced kinematically,
the user supplying positions taken by objects over time. An interesting
alternative is to simulate the motion dynamically, giving the objects phy-
sical mass properties and finding their motion under the influence of
forces and torques according to the dynamics equations of motion.
Dynamically-controlled motion is of particular interest for bodies with
many degrees of freedom, such as animals or flexible objects. This talk
will provide some background in setting up the dynamics equations of motion
for points, rigid bodies, articulated bodies, and flexible objects, and
then discuss issues of controlling such objects, including simulation of
collisions and constraints. Videotapes will be shown demonstrating
dynamically-animated bodies.
SYSTEMS SEMINAR, Thursday, October 15, 11 am, SF1101
Dr. Oscar M. Nierstrasz
Centre Universitaire d'Informatique
University of Geneva
"Active Objects"
Constructs for programming concurrent applications have traditionally
employed either shared memory or message-passing as the medium for communi-
cation and synchronization between concurrently executing processes.
Object-oriented languages can exploit both of these paradigms by modeling
objects as active entities communicating through remote procedure-calls.
We propose the additional mechanisms of delegation and delay queues to per-
mit interleaving and scheduling of concurrent activities in a manner that
is consistent with the operational interface of object specifications.
Higher-level constructs for defining concurrent subactivities and recover-
able atomic actions can be easily supported within this framework. The
mechanisms we propose are powerful enough to implement solutions to fami-
liar concurrency problems such as pipelining, constraint management and
``administration''.
NUMERICAL ANALYSIS SEMINAR, Thursday, October 15, 3:30 pm, GB119
Professor W.P. Tang
University of Waterloo
"Schwarz Splitting and Template Operators"
Schwarz alternating method (SAM) is a classical mathematical techniques
dated back by 1860. It was commonly known as an useful tool for theoretical
analysis but not for large scale scientific computations. Through our ana-
lyses, we show that some new development based on this idea is a competi-
tive solution technique for many important scientific applications, namely
an optimal complexity can be achieved for the model problems. These tech-
niques can be applied to parallel computations in a non--share memory
environment; composite grid and use of fast solver in an irregular region.
In order to identify for what kind of problems this new techniques are most
suitable, a new structure for the linear operator is presented. Some decay
results for sparse operators are shown. These results provide a theoretical
base of a successful application of these techniques.
SYSTEMS SEMINAR, Friday, October 16, 11 am SF1105
Dr. Satish K. Tripathi
University of Maryland
"Resource Allocation in Distributed Systems"
One of the major developments in computing in the last decade has been the
emergence of powerful workstations connected by local area networks (LAN).
The LAN provides services such as print servers and file servers to be
shared by attached workstations. In this talk we present static policies
for allocation of resources in such distributed systems.
Distributed systems are represented as network of queues. We consider pro-
duction queueing networks with multiple open and closed chains. Some of
the closed chains, which have a single customer each, require allocation of
resources in the network so as to maximize a weighted throughout perfor-
mance criterion. Chains with more than one customer can be decomposed into
many chains of one customer each. It is proved that an optimal allocation
of resources lies on a vertex of the set of feasible allocations. This
considerably reduces the search space for an optimal allocations. Applica-
tions of this result in distributed computing are discussed. Allocation of
files in LAN is presented as a special example.
THEORY SEMINAR, Friday, October 16, 2 pm, GB120
Mr. Mauricio Karchmer
Hebrew University
"A Monotone Depth Lower Bound
For Transitive Closure"
--
Jim Clarke -- Dept. of Computer Science, Univ. of Toronto, Canada M5S 1A4
(416) 978-4058
{allegra,cornell,decvax,linus,utzoo}!utcsri!clarke