Challenges and Beneficiations of differential privacy in transfer learning and federated learning
数学专题报告
报告题目(Title)：Challenges and Beneficiations of differential privacy in transfer learning and federated learning
报告人(Speaker)：吴慧雯 博士(之江实验室)
地点(Place)：后主楼 1223
时间(Time)：2023 年 04 月 13 日(周四), 10:00-11:00
邀请人(Inviter)：蔡永强
报告摘要
Differential privacy is a technique for safeguarding individual’s privacy while still allowing for valuable insights to be derived from data. It is widely employed in business intelligence and artificial intelligence due to its precise mathematical definition and theoretical findings. Nevertheless, implementing differential privacy necessitates careful deliberation of the trade-offs between data accuracy and privacy. Excessive noise can destabilize the computational system, whereas insufficient noise may result in privacy violations. In this presentation, we provide a concise overview of differential privacy, its applications in information transmission, and the potential obstacles. IJCAI’22 and WWW’22 accepted the corresponding papers.
主讲人简介
Dr. Huiwen Wu obtained a Ph.D. from the University of California, Irvine, in 2019. After that, she served as a senior algorithm engineer in Ant Group, focusing on designing and implementing privacy-preserving machine-learning algorithms. She entered Zhejiang Lab in 2022 as a Senior Researcher. Her main research interests are privacy-preserving machine learning algorithms, randomized optimization methods, and random matrix theory.
