ABSTRACTS OF TALKS
Presented by The Fields Institute and the Statistics Department, University
of Toronto
October 5th 1999
Probabilistic Networks and Expert Systems
Steffen Lauritzen, Department of Mathematics, Aalborg University
ABSTRACT
Bayesian and other probabilistic networks have now been established
as a basic tool for decision support under uncertainty. The theory and
methodology was developed in the late 80-s and exploits Markov properties
for directed acyclic graphs in combination with old and new algorithms
for local computation based on graph theory.
Microsoft Research has established a division to exploit such networks
in the computer technology. Hewlett-Packard has instead invested in
a research laboratory at Aalborg University and the Aalborg-company
HUGIN Expert A/S.
Bayesian networks are for example used for diagnosis of printer errors,
surveillance of steering rockets on the space shuttle, medical diagnosis
and numerous other applications, for example decoding in noisy communication
channels.
The lecture will attempt to convey important aspects of the theory
and methodology and the corresponding algorithms, together with a few
selected examples.
December 7th, 1999
Probabilistic and Statistical Techniques in Realistic Computer Graphics
Eugene Fiume, Department of Computer Science, University of Toronto
ABSTRACT
The use of probabilistic and statistical techniques has long had a place
in computer graphics. I will give a short visual motivation for the
use of these techniques in areas such as sampling and filtering for
improved image quality, the use of (quasi-)Monte Carlo techniques for
the computation of illumination integrals, and the use of stochastic
methods for optimisation. I will then motivate the use of stochastic
processes for the depiction of realistic natural phenomena such as smoke
and fire. Some recent animations illustrating these concepts will also
be shown.
February 1st 2000
Scaling Phenomena in Telecommunications
Murad Taqqu
Department of Mathematics, Boston University
ABSTRACT To appear.
March 7th, 2000
Two Timing: Politics and Response Latencies in a Bilingual Survey
Joseph F. Fletcher, Department of Political Science, University
of Toronto
ABSTRACT
A new addition to the survey researcher's toolkit is the recording of
how long it takes respondents to answer questions in an ordinary telephone
survey. I will highlight some of the statistical techniques used in
handling response time data as well as their substantive interpretation.
Results obtained in a four-wave bilingual panel survey, showing improvements
in the prediction of vote choice up to one year in advance of a federal
election by using response latency data will be featured.
These results are achieved with conventional computer-assisted telephone
interviewing (CATI) software, indicating that the immediate use of response
time measures is both practical and attractive for commercial as well
as academic survey units. Even so, response latencies are found to be
sensitive to political circumstance such that timings must be analyzed
separately for Anglophone and Francophone populations.
April 4th 2000
Smoothing techniques for surveys
James Stafford, Department of Public Health Sciences, University
of Toronto
ABSTRACT
The graphical display of survey data leads naturally to reducing it
to an aggregate form. For example, large data sets may be reported or
displayed as a histogram where raw data has been reduced to a small
number of bins and counts. Other examples include displays for scatterplot
data, binary data and longitudinal data.
One approach to adapting modern graphical techniques for survey data
is to simply apply them to this aggregate data. This may be the most
useful approach if the aggregation of survey data renders it in a from
for which modern graphical techniques were designed. These issues will
be discussed in a non-technical way.
