Robust Detection of Multiple Change-Points with High-dimensional Data
Grant Data
Project Title
Robust Detection of Multiple Change-Points with High-dimensional Data
Principal Investigator
Professor Yin, Guosheng
(Principal Investigator (PI))
Duration
36
Start Date
2016-07-01
Amount
488501
Conference Title
Robust Detection of Multiple Change-Points with High-dimensional Data
Presentation Title
Keywords
Categorical data, Change-point problem, High dimensionality, Latent variable, Penalized regression
Discipline
Probability & Statistics
Panel
Physical Sciences (P)
HKU Project Code
17326316
Grant Type
General Research Fund (GRF)
Funding Year
2016
Status
Completed
Objectives
1 To develop a latent variable approach for detection of multiple change-points with binary or polychotomous data, by casting the mean-change problem into a sparse regression model with a penalty or a suitable prior specification. 2 To propose a new homogeneity test for changes in a stream of multinomial data over time, via power enhancement and sequential moving-window monitoring. 3 To develop a robust nonparametric maximum likelihood approach to detecting structural changes in regression models using residuals from rank-based regression.