报告题目：A gentle introduction to distributed learning

报告人：张琼（中国人民大学）

报告时间：2024年11月8日（周五）晚6:30

腾讯会议（ID）：414583450

报告摘要：In recent years, distributed learning has emerged as a crucial approach for scaling machine learning models to large datasets. This lecture provides an introduction to distributed learning, covering key frameworks, including split-and-conquer methods and federated learning, which enable model training across decentralized devices without centralizing raw data—a critical feature for enhancing privacy. We'll discuss the risks associated with centralized data storage, particularly the potential for sensitive information leakage. Additionally, we'll explore some of the most prevalent attacks on machine learning systems, such as data poisoning, model inversion, and membership inference. By the end, participants will gain a foundational understanding of federated learning, its privacy-preserving benefits, and the challenges involved in securing distributed models against adversarial threats.

报告人简介：张琼，2015年本科毕业于中国科学技术大学少年班学院。2022年博士毕业于加拿大英属哥伦比亚大学统计系。2022年9月起加入中国人民大学统计与大数据研究院并担任助理教授。目前的研究兴趣包括:混合模型、分布式学习、联邦学习等。她的研究论文发表在Journal of Machine Learning Research, lEEE Transactions on Information Theory，ICCV等机器学习期刊和会议上，现主持国家自然科学基金青年项目。