Speaker: Cynthia Rudin, Massachusetts Institute of Technology

Title: Regulating Greed Over Time: An Important Lesson For Practical Recommender Systems

Abstract:

There is an important aspect of practical recommender systems that we noticed while competing in the ICML Exploration-Exploitation 3 data mining competition. The goal of the competition was to build a better recommender system for Yahoo!'s Front Page, which provides personalized new article recommendations. The main strategy we used was to carefully control the balance between exploiting good articles and exploring new ones in the multi-armed bandit setting. This strategy was based on our observation that there were clear trends over time in the click-through-rates of the articles. At certain times, we should explore new articles more often, and at certain times, we should reduce exploration and just show the best articles available. This led to dramatic performance improvements.

As it turns out, the observation we made in the Yahoo! data is in fact pervasive in settings where recommender systems are currently used. This observation is simply that certain times are more important than others for correct recommendations to be made. This affects the way exploration and exploitation (greed) should change in our algorithms over time. We thus formalize a setting where regulating greed over time can be provably beneficial. This is captured through regret bounds and leads to principled algorithms. The end result is a framework for bandit-style recommender systems in which certain times are more important than others for making a correct decision.

If time permits, I will discuss a separate project, which is an approach to decision tree (rule list) learning. This new method does not have the disadvantage of greedy splitting and pruning that haunts decision tree algorithms. It yields sparse logical models in a computationally efficient way. It is a fierce competitor for decision tree methods on a wide variety of problems, and it is more principled.The work on multi-armed bandits is joint work with Stefano Traca, Ed Su, and Ta Chiraphadhanakul.

The work on decision lists is joint with Hongyu Yang and Margo Seltzer.

Biography:

Cynthia Rudin is an associate professor of statistics at the Massachusetts Institute of Technology associated with the Computer Science and Artificial Intelligence Laboratory and the Sloan School of Management, and directs the Prediction Analysis Lab. Her interests are in machine learning, data mining, applied statistics, and knowledge discovery (Big Data). Her application areas are in energy grid reliability, healthcare, and computational criminology. Previously, Prof. Rudin was an associate research scientist at the Center for Computational Learning Systems at Columbia University, and prior to that, an NSF postdoctoral research fellow at NYU. She holds an undergraduate degree from the University at Buffalo where she received the College of Arts and Sciences Outstanding Senior Award in Sciences and Mathematics, and three separate outstanding senior awards from the departments of physics, music, and mathematics. She received a PhD in applied and computational mathematics from Princeton University. She is the recipient of the 2013 INFORMS Innovative Applications in Analytics Award, an NSF CAREER award, was named as one of the "Top 40 Under 40" by Poets and Quants in 2015, and was named by Businessinsider.com as one of the 12 most impressive professors at MIT in 2015. Her work has been featured in Businessweek, The Wall Street Journal, the New York Times, the Boston Globe, the Times of London, Fox News (Fox & Friends), the Toronto Star, WIRED Science, U.S. News and World Report, Slashdot, CIO magazine, Boston Public Radio, and on the cover of IEEE Computer. She is presently the chair of the INFORMS Data Mining Section, and currently serves on committees for DARPA, the National Academy of Sciences, the US Department of Justice, and the American Statistical Association.