Meeting Agenda

Wednesday, February 25, 1998

Location:

DC1304

Time:

1:30

Chair:

Wilkin Chau

1. Adoption of the Agenda - additions or deletions

2. Coffee Hour

Coffee hour this week:

???

Coffee hour next week:

???

3. Next meeting

Date:

Wednesday, March 4, 1998

Location:

DC1331 or Graphics Lab ??

Time:

1:30

Chair:

Blair Conrad

Technical presentation:

Richard Bartels

4. Forthcoming

Chairs:

1. Bill Cowan (Mar 11th)
2. Itai Danan (Mar 18th)
3. Ed Dengler (Mar 25th)

Tech Presenters:

1. Wilkin Chau (Mar 11th)
2. Blair Conrad (Mar 18th)
3. Bill Cowan (Mar 25th)

5. Technical Presentations

Presenter:

Pete Harwood

Title:

Scene Graph Rewriting - Some Results

Abstract:

Scene graph rewriting is a method for symbolically manipulating scene graphs. In this talk, I will be presenting some mildly interesting models I have created while experimenting with our scene graph rewriting system. I will first review the underlying ideas used to generate these scenes, which will include a simple example of code from our scene description language Grumble. I will then present a slide-show of the models, with some discussion.

6. General Discussion Items

• Blari: DGP visit

7. Action List

• Visiting prospective faculty members.
• Deadline for submission of Theses/Essays for Spring Convocation 1998 is May 30, 1998.
• Alias/SideFX talks coming; send abstracts to Mike McCool
• Blair Conrad coordinates U of T visit (with van@dgp)

8. Director's Meeting

9. Seminars

-----------------------------------------------------------------------


The infraNET Project Distinguished Speaker Seminar Series
---------------------------------------------------------

"THE NEXT GENERATION OF COMMUNITY NETWORKS"

By:    Don Cowan and Colin Mayfield
       University of Waterloo
Date:  Thursday, February 26, 1998
Time:  2:30 p.m. - 4:00 p.m.
Place: Davis Centre, Room 1302
       University of Waterloo

** There is no charge for this event.
** Pre-registering to reserve your seat is recommended.
** Seats are reserved on a first-come, first-serve basis.
** Register via our web site at:  http://infranet.uwaterloo.ca/
   or call 519-888-4567 ext.5611.

ABSTRACT
Community information covers a wide range of sources, and serves many
users, encompassing government, health and public safety organizations,
businesses of all types, service organizations, tourist attractions,
historical societies, and the general public. Communities are
repositories of large amounts of heterogeneous information, which need
to be searched, read, explored, acted upon, updated, and which offer
opportunities for collaboration and other forms of two-way and group
communication. The World-Wide Web, in conjunction with interactive map
interfaces and database systems, are suitable technologies for building
a community network to exploit this information. We will describe
prototype systems that have been created within local companies and the
University of Waterloo to support community networks, including one in
the Waterloo area.

Organizing information for a community has substantial impact on the way
we create and manage the underlying databases. In order for the
community network to prosper, community members and local businesses
must be actively involved in the acquisition and maintenance of
community information. We will explore this new model of community
networking, describe our experience with community networking in the
Waterloo area, and indicate potential new business opportunities.

-----------------------------------------------------------------------

DEPARTMENT OF COMPUTER SCIENCE
UNIVERSITY OF WATERLOO
SEMINAR ACTIVITIES

COMPUTER SCIENCE SEMINAR

                    -Monday, March 2, 1998

Brendan Frey, Beckman Institute for Advanced Science and
Technology, University of Illinois at Urbana-Champaign,
Illinois,  will  speak  on  ``Mixtures  of Orientation-
Adaptive  Low-Dimensional  Gaussians for Modeling High-
Dimensional Images''.

TIME:                10:30-11:30 a.m.

ROOM:                DC 1302

ABSTRACT

Recently, the "eigenface" technique (MIT) has proven to
be   a   successful   method  for  face  detection  and
recognition   in   multimedia   applications   such  as
annotated  video-conferencing.  In  this  technique,  a
high-dimensional   image   is  projected  onto  a  low-
dimensional    plane   in   order   to   simplify   the
classification problem. I am interested in an extension
of  eigenface  (principal components) analysis in which
the low-dimensional manifold is piece-wise linear. This
approach more accurately models local variations in the
data  due  to expression and pose, but at the same time
retains   the  tractability  of  modeling  with  linear
subspaces.

In   this   talk,  I  will  first  review  methods  for
dimensionality reduction (principal components analysis
and factor analysis) and clustering (K-means clustering
and  the  mixture  of  Gaussians).  These models can be
merged  to  produce  a  mixture of orientation-adaptive
low-dimensional Gaussians, (a.k.a. a "mixture of factor
analyzers").  This  model  can be fit to data using the
expectation  maximization  algorithm. I will present an
extension of this model that allows different Gaussians
to  share  directions  of  variation. This can act as a
regularizer  that prevents over-fitting of the training
data   and   captures   local   coherence  between  the
orientations  of  the  linear subspaces. I will present
experimental  results  on small problems throughout the
talk  to  illustrate  how  these algorithms work, and I
will  describe some recent work on applying these ideas
to image processing problems.

-----------------------------------------------------------------------

10. Lab Cleanup