ICES: Small: Auctions and Optimizations in Ad Exchanges

在互联网上浏览网页的浏览者会看到图片和视频等显示广告。传统上，网页发布商和广告商通过销售团队事先协商，决定展示哪些广告。一种新兴的购买和销售展示广告的方式是通过自动广告交易，即发布商和广告商通过实时拍卖交易广告印象的市场。该项目探讨了此类广告交换市场的计算、信息和经济方面。它将为这样的市场抽象出合适的模型，并研究基本的挑战，包括拍卖设计、在线优化、风险有限定价、实时竞价策略，甚至密码学方面的问题。具体的例子包括：在中介机构同时也是拍卖人的情况下，设计一个最优的拍卖，在这个层次结构的每一层决定哪一个中介机构来投标，并证明每一层的拍卖的完整性。解决这些问题需要经济学、金融学和优化学以及计算机科学的新方法、新概念和新工具。广告交换将影响几乎每一个互联网用户。这里描述的研究挑战的进展有可能直接影响这些系统和几乎每个互联网用户的体验。此外，面对技术挑战将汇集计算机科学家和经济学家，也推动这些学科解决非常高的性能挑战。例如，拍卖和优化解决方案必须在几十毫秒内工作，这是用户体验网页访问所需的时间。这就要求新的算法技术超越当前的艺术起点。最后，对信息在广告交换中的作用的详细分析——多少或多少信息与市场的运作相关——是互联网用户和最终社会非常感兴趣的。

Viewers who browse web pages on the Internet are shown display ads such as images and video. Traditionally, web page publishers and advertisers negotiate a priori through sales teams to determine which ads are shown. An emerging way to buy and sell display ads is via automated ad exchanges, which are marketplaces where publishers and advertisers trade ad impressions via real time auctions. This project explores computational, informational and economic aspects of such ad exchange markets. It will abstract suitable models for such markets and study the fundamental challenges, including problems in auction design, online optimization, risk-bounded pricing, real time bidding strategies, and even cryptography. Specific examples include the design of an optimal auction in the presence of a hierarchy of intermediaries who are also auctioneers, determining at each level of this hierarchy which intermediary to call for bids, and proving the integrity of auctions at each level of the hierarchy. Solving these problems requires new methods, concepts and tools from Economics, Finance and Optimization, and Computer Science. Ad exchanges will impact nearly every user on the Internet. Progress on research challenges described here has the potential to directly impact such systems and the experience of nearly every user on the Internet. Further, facing the technical challenges will bring together Computer Scientists and Economists, and also push these disciplines to address very high performance challenges. For example, auction and optimization solutions have to work in tens of milliseconds, the time it takes for users to experience a web page access. This calls for new algorithmic techniques beyond the current start of the art. Finally, a detailed analysis of the role of information in ad exchanges-how much or little information is relevant for the working of the marketplace-is of great interest to Internet users and ultimately the society.