报告题目： Fast Sketching Distributed PCA for Large-Scale Federated Data

报告时间：2026年8月18日下午15:30

报告地点：南湖校区办会楼二楼会议室1256室

主办单位：萝莉社

报告人：周兴才

报告人简介：周兴才，东南大学博士、东南大学控制科学与工程博士后，加拿大University of Alberta数学与统计系博士后，现为南京审计大学教授、博士生导师。曾主持“统计机器学习”为主题的国家自然科学基金面上项目和国家社会科学基金一般项目各1项；主持省部级项目3项；累计主持与参加国家级项目8项；在国际统计学顶刊JRSSB、JASA和人工智能顶刊TPAMI、顶会ICML以及权威期刊TNNLS、Techonometrics、JCGS、Statistica Sinica、EJS和中国科学: 数学等发表学术论文70余篇。任中国现场统计研究会生存分析分会副理事长、旅游大数据分会常务理事，江苏省应用统计学会理事。主要研究领域为统计机器学习和神经影像数据的因果推断。

摘要：We study distributed principal component analysis (PCA) for large-scale federated data when the sample size and dimension are both ultra-large. This type of data is currently very common, but faces numerous challenges in PCA learning, such as communication overhead and computational complexity. We develop a new algorithm FedFask (Fast Sketching for Federated learning) with lower communication cost overhead and lower computational complexityWe develop a new algorithm In FedFask, we adopt and develop technologies such as fast sketching, alignments with orthogonal Procrustes Fixing, and matrix Stiefel manifold via Kolmogorov-Nagumo-type average. Thus, FedFask has a higher accuracy, lower stochastic variation, and best representation of multiple randomly projected eigenspaces, and avoids the orthogonal ambiguity of eigenspaces. We show that FedFask achieves the same rate of learning as the centralized PCA uses all data, and tolerates more workers to parallel acceleration computation. We conduct extensive experiments to demonstrate the effectiveness of FedFask.