Online Learning for Advertising

Internet advertising is a multibillion dollar industry, and often provides the most significant source of revenue to companies providing other services on the Internet. Advertisers (seeking to buy impressions), and advertising networks (seeking to sell impressions) are faced with the task of solving large-scale strategic optimization problems with little human intervention.

In this talk I will describe how machine learning — specifically techniques from online learning and reinforcement learning — can be applied to solve various problems in this domain. On the side of the advertising networks, we will be interested in designing algorithms for ad-selection, as well as mechanisms for pricing advertisements in the face of strategic buyers. On the side of advertisers, we will consider bidding and budget-optimization problems. While I will focus on advertising as the motivating application, many of the models considered will generalize naturally to other settings, the closest being general (non-paid) content selection on the internet.