AF: Small: Information acquisition and revelation in games

长期以来，博弈论一直被用作建模和分析目标和偏好相互冲突的主体之间相互作用的结果的有力工具。最近博弈论与计算系统的融合促使人们开始研究信息在游戏中的作用。在高层次上，代理人在战略互动中的行为必然基于他们拥有的信息，以及他们对其他人持有的信息的信念。传统上，在经济学中，这些信息结构被认为是外生的；也就是说，信息结构是给定的，分析偏离了信息结构。除了少数例外，直到最近，研究才专注于将信息的获取和披露作为游戏本身的一部分。这里的一个关键动机是，信息的产生往往是以信息制造者的成本为代价的，这意味着信息的获取需要自私代理人的回报，这是一种需要博弈论观点的典型情况。同样，拥有信息的一个或多个代理人可能会发现，战略性地将部分或全部信息透露给其他代理人，以诱导特定的行为，使代理人或整个社会受益。在这两种情况下，设计最优或接近最优的信息获取或披露机制都是具有挑战性的，需要博弈论和算法洞察力的结合。本项目将分析信息获取和披露在游戏中的作用。具体地说，它将研究以下算法问题：(1)中央权威机构从自私代理人那里获得最优信息(在几种自然模型下)；以及(2)中央权威机构对自私代理人的最优信息揭示(在几种自然模型和信息结构下)。拟议的研究将有助于我们进一步理解信息在游戏中的作用，提供信息获取和揭示方面的基本特征。即将开发的算法也有可能影响在线拍卖的进行方式，在网上拍卖中，信息(或缺乏信息)发挥着核心作用。PI和共同PI都长期致力于积极传播结果和社区建设，而不仅仅是出版物。他们已经并将继续创建跨学科的研究生课程，并提供详细的课堂笔记，供其他教师使用。他们还将积极建立这一领域，在主要会议上提供教程，并为更广泛的受众编写概述文章。PI和共同PI致力于将本科生纳入研究工作，特别是努力招募女性和代表性不足的少数族裔。PI还将通过组织本地编程竞赛和显著加强南加州大学的计算机科学课程，继续帮助建立一个强大的计算机科学和问题解决社区。

Game theory has long been used as a powerful tool to model and analyze the outcomes of interactions between agents with conflicting goals and preferences. The recent confluence of game theory with computational systems has spurred research on the role of information in games. At a high level, agents' behavior in a strategic interaction will necessarily be based on the information they have available, as well as their beliefs about the information held by others. Traditionally, in economics, these information structures have been assumed to be exogenous; that is, the information structure is given, and the analysis departs from there.With few exceptions, it has only been recently that research has focused on the acquisition and revelation of information as part of the game itself. A key motivation here is that information is often produced at a cost to those who produce it, meaning that the acquisition of information requires rewards of selfish agents, a typical situation requiring game-theoretic viewpoints. Similarly, one or more agents possessing information may find it beneficial to strategically reveal some or all of it to other agents to induce particular behaviors, benefiting the agent, or society as a whole. In both cases, designing optimal or near-optimal mechanisms for acquiring or revealing information is challenging, and requires the combination of game-theoretic and algorithmic insights.This project will analyze the role of information acquisition and revelation in games. Specifically, it will study the following algorithmic problems:(1) Optimal information acquisition by a central authority from selfish agents (under several natural models); and(2) optimal information revelation by a central authority to selfish agents (under several natural models and information structures).The proposed research will help to further our understanding of the role of information in games, by providing fundamental characterizations of what is and is not achievable in terms of its acquisition and revelation. The algorithms to be developed also have the potential to impact the way online auctions are conducted, where information (or lack thereof) plays a central role.The PI and co-PI both have a long-standing commitment to active dissemination of results and community building going beyond mere publications. They have created and will continue to create interdisciplinary graduate courses and make detailed class notes available for use by other instructors. They will also actively establish the area by giving tutorials at major conferences and producing overview articles for a broader audience. The PI and co-PI are committed to including undergraduates in the research effort, with a particular effort to recruit women and underrepresented minorities. The PI will also continue to help build a strong computer science and problem solving community by organizing local programming contests and significantly strengthening the computer science curriculum at USC.