Academy of Mathematics and Systems Science, CAS
Colloquia & Seminars

Speaker:	王成杰，同济大学
Inviter:
Title:	Distributed Linear Regression with Bayesian Update Strategies
Language:	Chinese
Time & Venue:	2023.07.27 10:00-11:00 思源楼315
Abstract:	Traditional statistical methods encounter some difficulties in making inferences with large datasets when it is difficult to share data from different sources. Thus, distributed statistical problems have received a lot of attention in recent years. In this setup, datasets are partitioned over machines, which compute locally, and communicate short messages. Communication is often the bottleneck. We propose a multishot methods for distributed linear regression problems based on the idea of Bayesian inference. We do ridge regression on each machine whose prior is the weighted average result from the central server. The iterative results at the central server will converge to the weighted least squares estimate with little condition. This method requires only a small amount of data communication, and can be applied when the design matrix at each machine is not of full rank.