您所在的位置: 首页- 新闻公告- 学术讲座-


BDAI重点实验室研究生沙龙第36期:A Truthful Author-Assisted Paper Grading Mechanism for Maximizing the Conference Profit

BDAI 36 期.png


报告题目:A Truthful Author-Assisted Paper Grading Mechanism for Maximizing the Conference Profit




Abstract:Computer science related conferences have been growing rapidly in recent years and the number of submissions is exploding. However, reviewers usually do not spend enough time reading each article in detail and do not give accurate scores over papers even though they are supposed to. As a result, some high-quality papers are buried, and some low-quality papers are mixed into the proceedings, thus harming the profit of the conference. To elicit the true value of papers, we leverage the author’s information because we believe that the author is the person who understands the paper best. We study the paper admission problem from a perspective of mechanism design. In our model, the author knows the true value of the paper and only cares about whether his paper is accepted and the conference cares about the quality of accepted papers. We have designed a series of mechanisms to enhance the profit of the conference. Firstly, we design mechanisms for single-paper and show the optimal mechanism can be found efficiently. Due to the optimal mechanism is complicated to describe to the authors and thus is difficult to be adopted in practice, we introduce a simple mechanism called Two Menu Mechanism. Moreover, we propose the Union Mechanism for the author who has multiple submissions. It can incorporate multiple mechanisms designed for single-paper and is very easy to implement. Comparisons between different mechanisms are provided in both theory and experiment.

报告题目:Revenue Maximization Mechanisms for an Uninformed Mediator with Communication Abilities




Abstract: Consider a market where a seller owns an item for sale and a buyer wants to buy an item. Each player has a private type. It can be costly for them to come to an agreement on their own through communication. However, with a mediator as a trusted third party, the players can both communicate privately with the mediator and do not need to worry about leaking too much information. The mediator can design and commit to a multi-round communication protocol for both players. After each round, rational players will update their belief about the other player’s type, and decide whether to stay in the protocol for further communication or quit. The mediator is not able to force players to trade, but can influence their behavior by sending messages to them.

We study the problem of designing revenue-maximizing mechanisms for the mediator. We show that the mediator can, without loss of revenue, focus on the set of direct and truthful mechanisms. We formulate this problem as a mathematical program. Moreover, we give an optimal solution to the optimization problem in closed form. Our mechanism is also simple and has a threshold structure. Interestingly, we find that in the optimal mechanism, the mediator may even lose money in some cases.