Please use this identifier to cite or link to this item: https://scholarbank.nus.edu.sg/handle/10635/167559
Title: STRUCTURED DATA ANALYSIS: MODELS, ALGORITHMS AND THEORIES
Authors: ZHOU PAN
ORCID iD:   orcid.org/0000-0003-3400-8943
Keywords: structured data analysis, low-rank tensor analysis, constrained optimization, Riemannian optimization, deep learning theory, generalization theory
Issue Date: 22-Oct-2019
Citation: ZHOU PAN (2019-10-22). STRUCTURED DATA ANALYSIS: MODELS, ALGORITHMS AND THEORIES. ScholarBank@NUS Repository.
Abstract: With access to huge amounts of data, how to design effective models and efficient algorithms by leveraging intrinsic data structure for solving practical problems in computer vision and machine learning becomes very important. This thesis addresses this problem from three aspects: (1) designing effective structured data learning models, (2) developing efficient optimization algorithms, and (3) establishing solid theory to understand popular structured data learning models. Specifically, (1) this thesis designs two effective unsupervised structured data learning models for noisy tensor data recovery and clustering by leveraging the intrinsic low-rank structure in high dimensional tensor data, e.g., face images and videos. (2) It develops two kinds of efficient algorithms to solve the designed structured data analysis models, such as sparsity- or rank-constrained problems. (3) This thesis further establishes generalization and optimization performance guarantees for existing highly nonconvex deep learning models which are preferably used to analyze the complex structured data, e.g., the object images.
URI: https://scholarbank.nus.edu.sg/handle/10635/167559
Appears in Collections:Ph.D Theses (Open)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
PanZhou.pdf5.61 MBAdobe PDF

OPEN

NoneView/Download

Page view(s)

42
checked on Jun 11, 2021

Download(s)

9
checked on Jun 11, 2021

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.