Mathematics Graduate Students Organization

Seminar:

Date: March 1 (Note Date, Time, and Location Change!)in Milner 317

Every 15 minutes on Facebook, there are more photos uploaded than are present in the New York digital archive. Big data is ubiquitous in today’s society and it is only getting worse. While it is easy to drown in this flood of information, as mathematicians we are often confronted with the problem of taming this unwieldy beast. Every day, Google must rank webpages to answer search queries, Netflix must rank and suggest movies, etc. These million dollar ideas rely on sound mathematical theories and the problems underlying them are still an active area of research. How is this done and more generally how can one begin to sort, rank, and process such large amounts of information?

This introductory talk will use undergraduate linear algebra, calculus, and probability to examine the big data problem in the context of bracketology. Simply put bracketology is concerned with the problem: Given a league of teams playing a game and a history of past performance, can one predict future performance? We will analyze the Massey method which relies on simple linear algebra techniques. As an application, outcomes from the SEC football conference will be predicted (accurately!). While deceptively simple, this fast method is widely used and has consistently outperformed professional sports analysts. Time permitting, we will examine nonlinear extensions of this method. These extensions rely on the simple concept of maximal likelihood estimators and have been shown to perform in the 99-th percentile of the 2010 ESPN Challenge.