Local-Aggregate Model Paths for Massive Data via Distributed Optimization

• Activity Number: 611
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Yue Hu*+ and Genevera Allen
• Companies: Rice University and Rice University
• Keywords: massive structured data ; multi-subject neuroimaging ; distributed optimization ; solution path ; tensor data ; EEG

Abstract

Technological advances have led to a proliferation of massive structured data that is often collected and stored in a distributed manner. Examples include climate data, social networking data, crime incidence data, and biomedical imaging. We are motivated by the need to build predictive models for multi-subject neuroimaging data, based on each subject's brain imaging scans over time. Most existing methods ignore the spatial information due to computational demands. We propose a novel modeling and algorithmic strategy to apply generalized linear models (GLMs) to this massive 3D data with locations. Our method begins by fitting GLMs to each location separately, and then blends information across locations through regularization with what we term an aggregating penalty. Our so called, Local-Aggregate Models, can be fit in a completely distributed manner over the locations, and thus greatly reduce the computational burden. Furthermore, we propose to select the appropriate model via a novel type of solution path similar to regularization paths. We will demonstrate both the computational and predictive modeling advantages of our methods via simulations and an EEG classification problem.