CIS 700/004: Topics in Machine Learning and Econometrics
Department of Computer and Information Science
University of Pennsylvania
Instructor: Prof. Shivani Agarwal
Class Times/Venue: TR 1:30-3:00 pm, Towne 319 (First class meets on Tuesday January 17)
Machine learning is the computational and statistical science of learning predictive models from data. Econometrics is the branch of economics that uses statistics to model economic data. There is much synergy between the two fields, and increasing interest in topics at the intersection. This course will focus in particular on topics associated with analyzing and modeling rankings and choice data. Such rankings and choice data arise in several contexts, both economic and otherwise: examples include rankings of sports teams; consumer choices in transportation, housing, or energy markets; customer choices in retail markets (e.g. Amazon) or recommender systems (e.g. Netflix); voters' choices in elections; and many others. What types of mathematical and statistical models are appropriate for modeling such rankings and choice data? What types of machine learning methods can be used to automatically learn such models from data? How can we develop new learning methods that better take into account the characteristics and needs of different applications? This course will aim to answer some of these questions. Topics to be covered include:
We will also discuss related topics such as permutation models, binary choice probabilities, dueling bandits, noisy sorting in theoretical computer science, etc.
- Random utility models developed in statistics/mathematical psychology/econometrics;
- Discrete choice models developed in econometrics/marketing;
- Recent developments in machine learning for modeling rankings and choice data.
The first few weeks will consist of lectures by the instructor giving an overview of several of the topics above. The remainder of the course will involve reading/presenting papers on these topics and working in small teams on a (potentially research-level) project.
The ideal student taking this course will have taken CIS 520 (or equivalent course elsewhere) and will have a reasonable degree of mathematical maturity.
Students who have taken STAT 520/521 and are willing to pick up some machine learning tools as needed are also welcome. Auditors/postdocs are welcome to join, but if you come to more than two lectures, you will be expected to lead at least one paper discussion in the class!
Note: If you have taken CIS 519 and wish to take this course, you may have to cover some ground on your own.
This course can be used toward fulfilling the CIS/PhD seminar requirement.