Wednesday, April 02, 2014, 02:00pm
"Machine Learning in an Exchange Environment"
Abstract: Machine learning involves very large regressions, with as many as a billion explanatory variables. Due to the scale, point estimates are typically used, so that common economic concepts like standard errors are ignored. The talk focuses on applications of ML in an exchange environment, as arises in internet advertising. The use of ML in an auction suggests several ways of improving ML: dealing with the winner's curse, accommodating the inevitable prediction errors into pricing, and choosing a loss function appropriate to the application. In addition, it is shown that active learning strategies have modest payoffs.
Bio: R. Preston McAfee received his undergraduate degree in economics from the University of Florida, and M.S. in mathematics and a Ph.D in economics from Purdue University. McAfee is a research director at Google. Previously he was chief economist of Yahoo! and has served as Professor at Caltech, U of Texas, and U of Chicago. He has served as a co-editor of the American Economic Review, editor of Economic Inquiry and currently is a co-editor of ACM-TEAC. McAfee has written extensively on auctions, pricing and antitrust and was one of the designers of the US PCS auctions. He has run auctions in Mexico and advised governments on auction design. In addition, he has testified in high profile antitrust cases including the FTC v. Rambus, and the mergers Exxon and Mobil, BP and Arco, Peoplesoft and Oracle, and has testified before three United States Senate committees on antitrust enforcement and gasoline pricing.
Hosted by: Jason Hartline and EECS