A Google TechTalk, presented by Jason Liang, 2023-02-14
ABSTRACT: In digital advertising, advertisers procure ad impressions simultaneously on multiple platforms, or so-called channels. Advertisers communicate their procurement goals to channels, where the channels then consequently run automated bidding algorithms to participate in ad auctions on their behalf.

In the first part of this talk, we address a single advertiser’s conversion (e.g. ad clicks) maximization problem while satisfying aggregate return-on-investment (ROI) and budget constraints across all channels. In practice, an advertiser does not have control over, and thus cannot globally optimize, which individual ad auctions she participates in for each channel. Instead, the advertiser authorizes each channel to procure impressions on her behalf: the advertiser can only utilize two levers on each channel, namely setting a per-channel budget and per-channel target ROI. We analyze the effectiveness of each of these levers for solving the advertiser’s global multi-channel problem. Further, we present an efficient learning algorithm to optimize over per-channel levers to approximate the global optimal conversion.

In the second part of this talk, we address a single channel’s problem to improve individual fairness for auction outcomes. Channels have commonly focused on maximizing total advertiser value (or welfare) to attract advertiser traffic and spend, and have resorted to machine learning predictions on advertiser values (also known as machine-learned advice) to improve ad auction designs. Yet, such improvements could come at the cost of individual bidders’ welfare, consequently eroding fairness of a channel, and do not shed light on how particular bidding strategies impact individual fairness. Motivated by this, we present a novel fairness metric that measures an individual bidder’s welfare loss, and also uncovers how advertiser strategies relate to such losses. Under this metric, we then study how channels can utilize ML advice to improve welfare guarantees and fairness on the individual bidder level. We motivate a simple approach that directly sets such advice as personalized reserve prices, and present individual welfare guarantees for classic VCG, GSP and GFP auctions under such ML-advice-based reserves.

Bio: Jason Liang is a 5th year PhD at the MIT Operations Research Center, advised by Negin Golrezaei and Patrick Jaillet. He graduated summa cum laude from Columbia University with a B.S. in Operations Research. Jason’s research interests include online learning, mechanism design, and revenue management for online marketplaces. Jason was also a student researcher in the Market Algorithms Group at Google hosted by Vahab Mirrokni and Yuan Deng during 2022 winter.

A Google Research Algorithms Seminar.

свернуть