演讲人:张磊夫 助理教授 香港科技大学(广州)
时间:2024年6月18日(周二)12:00-13:00
地点:同济大厦A楼505室
同步#腾讯会议:599834467
会议密码:249567
链接:https://meeting.tencent.com/dm/O8YJuUbIQLhD
摘要:I consider a decentralized oracle where an adversary controls part of the nodes, and the remaining nodes are strategic in maximizing their expected payoffs. I show that a robust compensation mechanism—under that mechanism, there is an equilibrium in which truthful reporting is the unique optimal response for strategic nodes regardless of the adversary’s strategy—generally does not exist. Moreover, I stress that the popular aggregating method for obtaining a robust consensus may fail even without an adversary and ignores the intrinsically high-dimensional structure of decentralized oracles. I show that adversarial nodes which look normal in every single dimension could be detected from a “global” view by utilizing state-of-the-art algorithms in machine learning.
简介:张磊夫博士现为香港科技大学(广州)社会枢纽金融科技学域助理教授。他是一位研究兴趣广泛的金融经济学者,尤其关注金融学与计算机科学的交叉领域。在加入香港科技大学(广州)之前,他在芝加哥大学布斯商学院从事博士后研究。张磊夫博士于圣路易斯华盛顿大学获得金融学博士学位,并在该校计算机科学与工程系修读了多门课程。此外,他还曾在芝加哥大学和丰田工业大学芝加哥分校(TTIC)旁听过部分计算机科学相关课程。张博士本科就读于北京大学,后获全额奖学金在香港中文大学攻读经济学研究型硕士。