Methods For High Dimensional Matrix Computation And Diagnostics Of Distributed System
Files
No Access Until
Permanent Link(s)
Collections
Other Titles
Author(s)
Abstract
Big data provides opportunities, but also brings new challenges to modern scientific computing. In this thesis, we conduct sparse principal component analysis (SPCA) on high dimensional matrices. We propose a modified curvilinear algorithm to solve eigenvalue optimization with orthogonal constraints, and combine it with an augmented Lagrangian method to improve its computational efficiency. We compare our algorithm against standard PCA on the recovery of low-rank tensors and a mean-reverted statistical arbitrage strategy. The explosion of big data has also influenced the development on distributed computing systems. For debugging purposes, we are interested in predicting server run-time based on system data early in the process. We study discriminative models in functional data analysis, and introduce generative models that capture server regime-change behaviors. We also design computational methods, including a blocked Gibbs sampler, to improve the accuracy and efficiency of model estimation.
Journal / Series
Volume & Issue
Description
Sponsorship
Date Issued
Publisher
Keywords
Location
Effective Date
Expiration Date
Sector
Employer
Union
Union Local
NAICS
Number of Workers
Committee Chair
Committee Co-Chair
Committee Member
Turnbull, Bruce William