13
Nov
KR Seminar
There will be pizza!
——————————
————————————–
Date: Wednesday, Nov 16, 2011
Time: 12:00pm–2:00pm
Room: PT266
——————————————————————–
Date: Wednesday, Nov 16, 2011
Time: 12:00pm–2:00pm
Room: PT266
——————————————————————–
Title: Decision-theoretic preference elicitation using
generalized additive utilities
Author: Darius Braziunas
Abstract:
In this talk, I will present our work (with Prof. Craig Boutilier) on
decision-theoretic approach to preference elicitation with generalized
additive (GAI) utilities.
09
Nov
Machine Learning Seminar
George will speak about some work-in-progress on training RBMs with
high dimensional multinomial outputs. The details are as follows:
Date: Thurs Nov 10, 11AM, Pratt 290C
Speaker: George Dahl http://www.cs.toronto.edu/~gdahl/
Title: Training Restricted Boltzmann Machines on Word Observations.
Abstract:
I will give a talk discussing preliminary results from ongoing work
with Ryan Adams and Hugo Larochelle.
The restricted Bolzmann machine (RBM) has been used in many different
modeling contexts, typically characterized by varying types of
observation vectors, e.g., binary values, integers and real numbers.
One type of observation that has always proven problematic is the
high-dimensional multinomial, which would be useful for modeling
natural language data with large vocabularies. Conventional learning
methods, such as contrastive divergence have not succeeded in this
setting, as it requires softmax units that can generate hundreds of
thousands of discrete outcomes, matching the size of the vocabulary
being modeled. Even generating data from such a distribution requires
linear time, making it unwieldy to produce negative examples for
learning. I will be discussing (preliminary) work where we address
this issue by using a more general class of Markov chain Monte Carlo
moves on these visible units. With the use of appropriately-selected
transition kernels, it is possible to sample (correlated) negative
observations in constant time, making it viable to learn with
contrastive divergence. We confirm the success of this approach by
showing that RBMs can be used to learn meaningful word representations
on hundreds of millions of n-gram word windows.
York University-MITACS Vision Science Summer School
The Centre for Vision Research (CVR) at York University offers a one-week, all-expenses-paid undergraduate summer school on vision science. This year’s program will be held June 3-8, 2012.
The program includes talks by CVR faculty members on current research topics in vision science, as well as hands-on projects in CVR laboratories. The curriculum reflects the wide range of research areas at CVR, which includes basic research on vision in humans and animals, covering both neuroscience and behaviour, as well as applied topics such as computer vision, virtual reality, visual perception in low-gravity environments, and vision in clinical populations. For an idea of the topics to be covered, see last year’s schedule at www.yorku.ca/cvrss/schedule2011.pdf.
The program will accept 24 undergraduate students who are interested in pursuing a career in scientific research. It is intended mainly for students who are planning to apply to graduate school in the fall of 2012 for admission in 2013, and who are interested in investigating vision science as a possible area of research. Both Canadian and international students are encouraged to apply.
The program provides on-campus accommodations, breakfast and lunch each day, a closing banquet, and reimbursement for transportation costs.
Application instructions are available on the summer school website, www.yorku.ca/cvrss. The application deadline is February 1, 2012, and applicants will be notified of decisions by March 1, 2012.
For further information, see the summer school website, www.yorku.ca/cvrss, or write to Dr. Joseph DeSouza (desouza@yorku.ca).
This program is funded by York University and MITACS.
Poster available at www.yorku.ca/cvrss/poster2012.pdf.
09
Nov
Vision Seminar
On Friday next week (Nov 11), at 11am in Pratt 266, Marcus Brubaker will present a talk entitled “Physics in Human Motion Estimation and Scene Understanding”.
For the abstract of the upcoming talk, or to see a list of other future vision seminars, see http://www.cs.toronto.edu/vis/seminars/index.html
25
Oct
Machine Learning Seminar
Date: Thurs, Oct 27 2011, 11AM, Pratt 290C
Speaker: Kevin Swersky
Title: On Autoencoders and Score Matching for Latent Energy-Based Models
Abstract:
We consider estimation methods for the class of continuous-data energy
based models (EBMs) with latent variables, for the purpose of
unsupervised feature learning. Our main result shows that estimating
the parameters of an EBM using score matching when the conditional
distribution over the visible units is Gaussian corresponds to
training a particular form of regularized autoencoder. This provides
deep links between these two families of models. We show how different
Gaussian EBMs lead to autoencoder architectures that incorporate
higher-order features that model covariance information. Our
experiments suggest that regularized higher-order autoencoders are
capable of learning competitive features for classification compared
to higher-order EBMs, while allowing for the usage of black-box
optimization tools to learn the parameters.
This is joint work with Marc’Aurelio Ranzato, David Buchman, Benjamin
Marlin and Nando de Freitas.
25
Oct
Vision Seminar
Friday this week (Oct 28), at 12 noon in Pratt 266, Mohammad Norouzi will present his research on “Efficient near neighbor search on similarity-preserving binary codes”
Please note the time of the talk, as it begins an hour later than normal ie., at noon)
For the abstract of the talk, or to see a list of upcoming vision seminars,
see http://www.cs.toronto.edu/vis/seminars/index.html
25
Oct
KR Seminar
————————————–
Date: Wednesday, Nov 2, 2011
Time: 12:00pm–2:00pm
Room: PT266
——————————————————————–
Date: Wednesday, Nov 2, 2011
Time: 12:00pm–2:00pm
Room: PT266
——————————————————————–
Title: StarPlayer
Author: Timothy Joseph Cerexhe
Abstract:
General game playing (GGP) attempts to focus game playing research on
techniques with broader applications than a single game. Currently most of
this effort is devoted to games with perfect information.
There is increasing interest in bridging this competition into imperfect
information games, however there are significant computational and
algorithmic barriers to this transition.
Our group at UNSW has produced one of the first imperfect information GGP
players, which came second in the (very) recent first German Open.
This talk will introduce the competition, the approach we took to produce
the UNSW StarPlayer, and some future directions.
Bio:
Timothy Cerexhe is a second-year PhD student at the University of New
South Wales (UNSW) in Sydney, Australia.
His research interests are in cognitive robotics and general game playing.
He is working towards a thesis on analysing and targeting special cases of
action theory, with applications to planning and game playing.
13
Mar
Faculty Lecture Series
|
Event date: Tuesday, March 22, 2011, at 11:00 AM
|
|
Location: Bahen Centre for Information Technology, 40 St. George Street, Rm. 1180
|
|
Ron Baecker, Professor, Human Computer Interaction, Department of Computer Science, University of Toronto
Biography: Ron Baecker is the Bell Chair in Human-Computer Interaction, co-founder of the Dynamic Graphics Project, founder of the Knowledge Media Design Institute, and founder and director of the Technologies for Aging Gracefully lab (TAGlab) at the University of Toronto. He is also Affiliate Scientist with the Kunin-Lunenfeld Applied Research Unit of Baycrest and Adjunct Scientist with the Toronto Rehabilitation Institute. He has been named one of the 60 Pioneers of Computer Graphics by ACM SIGGRAPH, has been elected to the CHI (Computers and Human Interaction) Academy by ACM SIGCHI, and has been given the Canadian Human Computer Communications Society Achievement Award. His B.Sc., M.Sc., and Ph.D. are from the Massachusetts Institute of Technology.
C.C. (Kelly) Gotlieb, Professor Emeritus, Department of Computer Science, University of Toronto Chiefly about Computing Monday, March 28, 2011 4:00 pm, Bahen Centre for Information Technology, 40 St. George Street, Rm. 1160 Kelly Gotlieb is the founder of the Department of Computer Science at the University of Toronto, and is widely regarded as the “Father of Computing in Canada”. As stated in his 1996 Order of Canada investiture, “(Kelly) has been largely responsible for leading Canadians into the modern age of computing.” Now in his 90th year, Gotlieb remains extremely active. He has served for the last twenty years as the co-chair of the awards committee for the Association of Computing Machinery (ACM), which is the largest international scientific and educational organization for computing, and continues to this day to give invited talks at various meetings. Awards and honors include: Fellowships in the Royal Society of Canada, the Association of Computing Machinery, the British Computer Society, and the Canadian Information Processing Society; and four honorary doctorates from the University of Toronto, the University of Waterloo, the Technical University of Nova Scotia and the University of Victoria.
Craig Boutilier, Professor, Artificial Intelligence, Department of Computer Science, University of Toronto Individual, Social and Market Choice: Computational Assessment of Human Preference Tuesday, May 3, 2011 11:00 am, Bahen Centre for Information Technology, 40 St. George Street, Rm. 1180 Craig Boutilier served as Chair of the Department of Computer Science from 2004-2010. He has provided consulting, visiting or advisory roles at Rockwell International, Stanford University, Brown University, Carnegie Mellon University, and CombineNet, Inc. He recently served as Program Chair for 21st International Joint Conference on Artificial Intelligence (IJCAI-09), and is Associate Editor-in-Chief of the Journal of Artificial Intelligence Research. Boutilier is a Fellow of the Association for the Advancement of Artificial Intelligence (AAAI), has been awarded the Isaac Walton Killam Research Fellowship, an IBM Faculty Award and a Killam Teaching Award from the University of British Columbia. He received his Ph.D from the University of Toronto and joined the University of British Columbia for eight years; returning to Toronto in 1999. |
The Department of Computer Science, University of Toronto will hold its 5th annual
Research in Action Showcase
Friday, April 8th 2011
11:00 am – 3:00 pm
Bahen Centre for Information Technology
40 St. George Street, Toronto, ON
A showcase of the innovative and diverse research of the Department of Computer Science, University of Toronto.