A Statistical Approach to High-dimension High-order Spatial-Temporal Data Clustering

Professor Feng, Long   (Principal Investigator (PI))

36

2023-01-01

655390

A Statistical Approach to High-dimension High-order Spatial-Temporal Data Clustering

1) Spatial-temporal data 2) Clustering 3) Tensor decomposition 4) High-dimensional data 5) Sparse SVD

Probability & Statistics

Physical Sciences (P)

21313922

Early Career Scheme (ECS) 2022/23

2022

On-going

 1  For both matrices and tensors represented spatial-temporal data, develop a unified framework for clustering and spatial regions/temporal periods detection.  2  Develop efficient algorithms for tensor decomposition. The tensor decomposition algorithm will be used to address the high-dimension high-order spatial-temporal data clustering problem.  3  Provide theoretical guarantees for recovering sparse tensor with CP low-rank structure. In particular, convergence results of the alternatively updating algorithm will be developed.  4  Provide theoretical guarantees for the clustering accuracy and region detection consistency.  5  Evaluate the proposed method in a comprehensive simulation study to demonstrate its region detection and clustering accuracy for high-dimension high-order spatial-temporal data.  6  Apply the proposed methods to real fMRI data, using the UK biobank database. Collaborate with general health researchers and validate the detected brain regions associated with various intellectual disabilities.  7  Develop statistical packages for the proposed methods and facilitate psychiatry researchers in analyzing brain imaging data, particularly, MRI and fMRI data.