Daniel TarlowMachine Learning Research GroupUniversity of Toronto Department of Computer Science dtarlow cs toronto edu |
|
I am broadly interested in the representation, reasoning, and learning needed to apply probabilistic models to hard real world problems. I spend most of my time working on MAP inference problems -- specifically, new formulations and applications of max-product belief propagation and related decomposition-based algorithms. Beyond this, I work a bit on nonparametric Bayesian models, clustering, image segmentation, 3D reconstruction from feature trajectories, recommendation systems, and other matrix-factorization-based models.
Last year, I was on leave in San Francisco, splitting my time between modeling hurricane damage to oil platforms in the Gulf of Mexico and learning about the physics and uncertainty associated with modeling how we use energy in cities and buildings.
Machine Learning & Computer Vision
Learning Articulated Skeletons From Motion (2007)
David Ross,
Daniel Tarlow, and
Richard Zemel.
Workshop on Dynamical Vision at International Conference on Computer Vision (WDV-ICCV).
[pdf]
[webpage]
[bibtex]
Using Combinatorial Optimization within Max-Product Belief Propagation (2007)
John Duchi,
Daniel Tarlow,
Gal Elidan, and
Daphne Koller.
Advances in Neural Information Processing Systems (NIPS 19).
[pdf]
[bibtex]
Civil & Environmental Engineering
Automatically Calibrating a Probabilistic Graphical Model of Building Energy Consumption
IBPSA Conference on Building Simulation, Glasgow, Scotland, Summer 2009.
Flexible Priors for Exemplar-based Clustering
Uncertainty in Artificial Intelligence, Helsinki, Finland,
Summer 2008.
Toronto Machine Learning Group Seminar,
Toronto, Canada, Winter 2008.
[pdf]
[video]
Learning Articulated Skeletons from Motion
CIFAR Summer School on Neural Computation and Adaptive Perception,
Summer 2007.
[pdf]
Using Combinatorial Optimization within Max-Product Belief Propagation
Toronto Machine Learning Group Seminar, Toronto, Canada,
Fall 2006.
[ppt] [pdf]
The Role of Features in a Feedback-based Ranking System
Google TechTalk,
Mountain View, California, Summer 2006.
Partition-based Inference in Markov Networks
with John Duchi.
DAGS,
Stanford, California, Spring 2006.
Learning in General Games
with Lee Zen and Ankit Garg.
Stanford Logic Group,
Stanford, California, Winter 2005.
[link]
Automated Grading of Logic-based Homework Problems
Stanford Logic Group,
Stanford, California, Summer 2004.