Including talks, posters, theses and installations.
|Matthew Hoffman, Eric Brochu and Nando de Freitas. 2011. Portfolio Allocation for Bayesian Optimization. 27th Conference on Uncertainty in Artificial Intelligence (UAI2011).
An updated and expanded version of our 2010 arXiv paper (below), which was also the basis of Chapter 5 of my PhD thesis. How to solve the problem of deciding the acquisition (infill) function for Bayesian optimization when the objective is poorly-understood. The approach here is to use a portfolio of acquisition functions which automatically trade off by performance.
|Kenji Okuma, Eric Brochu, David G. Lowe and James J. Little. 2011. An Adaptive Interface for Active Localization. International Conference on Computer Vision Theory and
Description of an application for using active learning to assist localizing and labelling objects in images. We present an approach for reducing the number of labelled training instances required to train an object classifier and for assisting the user in specifying optimal object location windows. Our active learning system provides a mean performance improvement of 4.5% in the average precision over a state of the art detector on the PASCAL Visual Object Classes Challenge 2007 with an average of just 40 minutes of human labelling effort per class.
|Eric Brochu. Interactive Bayesian Optimization: Learning Parameters for Graphics and Animation. PhD thesis. University of British Columbia, December 2010.
|Eric Brochu, Vlad M. Cora and Nando de Freitas. A Tutorial on Bayesian Optimization of Expensive Cost Functions, with Application to Active User Modeling and Hierarchical Reinforcement Learning. eprint arXiv:1012.2599, arXiv.org, December 2010.
A tutorial on using Bayesian optimization for a variety of real-world applications. Follows the same framework as my PhD thesis, but written for a more general audience.
|Eric Brochu, Tyson Brochu and Nando de Freitas. A Bayesian Interactive Optimization Approach to Procedural Animation Design. ACM SIGGRAPH/Eurographics Symposium on Computer Animation, 2010.
Geared towards a graphics audience, this paper introduces some new techniques in applying Bayesian optimization to the problem of finding parameters for procedural animation, and presents a working application using those techniques.
There is also a video which explains the whole thing in under five minutes with no math.
|Eric Brochu, Matt Hoffman and Nando de Freitas. Hedging Strategies for Bayesian Optimization. eprint arXiv:1009.5419, arXiv.org, September 2010.
How to solve the problem of deciding the acquisition (infill) function for Bayesian optimization when the objective is poorly-understood. The approach here is to use a portfolio of acquisition functions which automatically trade off by performance.
|Eric Brochu, Vlad M. Cora and Nando de Freitas. 2009. A Tutorial on Bayesian Optimization of Expensive Cost Functions, with Application to Active User Modeling and Hierarchical Reinforcement Learning. Technical Report TR-2009-023. University of British Columbia, Department of Computer Science.
Earlier version of the 2010 tutorial (above, with same title). Please use the arXiv version, this one is obsolete!
|Ruben Martinez-Cantin, Nando de Freitas, Eric Brochu, Jose Castellanos and Arnaud Doucet. 2009. A Bayesian Exploration-Exploitation Approach for Optimal Online Sensing and Planning with a Visually Guided Mobile Robot. Autonomous Robots, 27(2):93–103.||2008|
|Nando de Freitas, Ali Davar, Eric Brochu, Mike Klaas and Kevin Leyton-Brown. 2008. Worio: A Web-Scale Machine Learning System. Advances in Neural Information Processing Systems 21, Demonstration Session.
From late 2007 to early 2009, after finishing my thesis proposal, I took an extended leave to work with Worio, a Vancouver startup. This was a demo of that project, primary for a Machine Learning audience.
|Ali Davar, Mike Klaas and Eric Brochu. 2008. US Patent 20070156615 entitled Method for training a classifier.
My first (and so far only) patent, also from my time with Worio.
|Eric Brochu, Nando de Freitas and Abhijeet Ghosh. 2007. Active Preference Learning with Discrete Choice Data. Advances in Neural Information Processing Systems 20.
This paper covers the details of the model used in our SIGGRAPH poster.
|Eric Brochu, Abhijeet Ghosh and Nando de Freitas. 2007. Preference Galleries for Material Design. ACM SIGGRAPH Sketch.
Awarded First Place at the ACM SIGGRAPH Student Research Competition.
|Eric Brochu. 2006. VAGUE: A Multimedia Navigation Tool. Multimedia Installation, Vancouver Art Gallery.
A software installation I did for the Vancouver Art Gallery exhibit “Emily Carr: New Perspectives on a Canadian Icon”. The software uses Machine Learning techniques to organize and manipulate images interactively. A physical installation appeared in the lobby of the gallery October 2006 to January 2007.
(Note that the news report linked to gets some details wrong, crediting “Simon Fraser University students”. I’m at UBC, and I’m reasonably confident I’m only one person.)
|Eric Brochu, Nando de Freitas and Kejie Bao. 2004. Owed to a Martingale: A Fast Bayesian On-Line Algorithm for Multinomial Models. UBC Technical Report TR-2004-8.
A fast and powerful EM algorithm for multinomial data. The theory is incomplete so far, which is why it hasn’t gone past the tech report stage.
|Eric Brochu. 2004. Frail and Realistic: Learning from Humans. Simulation and Other Re-enactments: Modeling the Unseen. Banff New Media Institute summit.
I gave this talk based on my thesis at a summit of artists and scientists. I wasn’t really prepared and they put me on a panel with Oscar-winning computer scientist Ken Perlin! Pretty intimidating, and I was pretty nervous during my talk, but a lot of people seemed interested and talked to me about my work afterward, which was great.
|Eric Brochu. 2004. milq. Master’s thesis. University of British Columbia.
My MSc thesis. Describes a novel learning technique for predicting the emotional qualities of music by combining audio signal analysis with statistical information on the patterns in which different adjectives are used to describe music. Also, the shortest thesis title in the history of the department.
|Asher Lipson, Hendrik Kueck, Eric Brochu and Nando de Freitas. Machine Learning for Computer Games. First International Digital Games Research Conference (DiGRA 2003).
Poster presentation and demo.
|Asher Lipson, Eric Brochu and Nando de Freitas. 2003. The ‘Touring’ Test: Human-Like Play in Computer Games. Snowbird Learning Workshop (Learning@Snowbird 2003).
Refereed poster presentation.
|Eric Brochu, Nando de Freitas and Kejie Bao. 2003. The Sound of an Album Cover: Probabilistic Multimedia and IR. Ninth Annual Workshop on Artificial Intelligence and Statistics (AI-Stats 2003).
Builds on my NIPS*2002 paper, describing a probabilistic database that combines music, images and text, and can be searched on any combination of those. Selected for oral presentation at AI-Stats 2003.
|Nando de Freitas, Eric Brochu, Kobus Barnard, Pinar Duygulu, and David Forsyth. 2003. Bayesian Models for Massive Multimedia Databases: A New Frontier. UBC Technical Report TR-2003-005.
This work was presented at 7th Valencia international meeting on Bayesian Statistics, 2002.
Eric Brochu and Nando de Freitas. 2003. “Name That Song!”: A Probabilistic Approach to Querying on Music and Text. Advances in Neural Information Processing Systems 15.
My first refereed publication, describing a probabilistic mixed-media database and search. My AI-Stats paper is probably more interesting, though, as the research was a bit further along by then.