Tyler Lu

 

Knowledge Representation and

Machine Learning Group

Department of Computer Science

University of Toronto

10 King's College Road

Toronto, Ontario,

Canada M5S 3G4

Email: tl@cs.toronto.edu

Hello and welcome to my homepage!


I am a PhD student in the Department of Computer Science at the University of Toronto, advised by Craig Boutilier, and affiliated with the Knowledge Representation and Machine Learning groups. I am interested in solving problems that will have a significant positive influence on the world.


My doctoral research centers around group decision making (known as social choice in economics) in the presence of incomplete preferences. Applications of my research are ubiquitous in everyday life. For example, in low-stake decision support where groups of friends need to reach consensus on restaurant choices, vacation destinations, etc., or in high-stake settings, such as elections, and in corporate environments where management might be interested in aggregating opinions from employees.  Our recent IJCAI-11 paper on robust approximation in social choice has allowed for agents (users, employees) to reveal only partial preferences, such as top few choices, while our algorithms can aggregate such preferences and recommend a group choice with provable guarantees on quality (using the concept of minimax regret). Further, if not enough preference is specified to make a good decision ("good" can be user-defined) our algorithms can intelligently elicit the necessary preferences that will lead to a good decision. Our empirical results show that, contrary to worst-case predictions, real preference data exhibit enough patterns that only a fraction of preference information is needed to make optimal or near optimal decisions.


Because real preferences often have structure, such as similar groups of users, we have been developing (see our ICML-11 paper which, to the best of our knowledge, is the first to learn with pairwise comparisons under non-trivial rank distributions) and applying probabilistic tools from machine learning and statistics to model individual preferences. Such models allow for a variety of principled and informed group (or even personalized) decision-making that fit in a minimax or Bayesian paradigm. See our recent discussion paper in IJCAI-11 social choice workshop.


Before entering doctoral studies, I received my undergraduate and masters (with Shai Ben-David) degrees at the University of Waterloo.