B0583
Title: Renewable maximum likelihood estimation in generalized linear models for streaming data
Authors: Lan Luo - University of Iowa (United States)
Peter Song - University of Michigan (United States) [presenting]
Abstract: A new incremental learning algorithm is presented to analyze streaming data using the generalized linear models. The proposed method is developed within a new framework of renewable estimation, in which the maximum likelihood estimation can be renewed with current data and summary statistics of historic data, but with no use of any historic data themselves. In the implementation, we design a new data flow, called the rho architecture to accommodate the data storage of current and historic data, as well as to communicate with the computing layer of the system in order to facilitate sequential learning. We prove both estimation consistency and asymptotic normality of the renewable MLE, and propose some sequential inferences for model parameters. We illustrate our methods by various numerical examples from both simulation experiments and real-world analysis.
