The aim of this dissertation is to explore the deep architectures via stacked autoencoders (AE) and to enhance their performance in the regression analysis and the classification task. To this end, two major real-world problems are considered, one with real-valued data for the regression analysis and one with nominal data for the classification problem.
While there is abundant research conducted on images (nominal data) using deep architectures, studies on regression analysis using real-valued data were less available at the time this research was started. Therefore, one of the motivating elements of my dissertation has been the lack of research in regression analysis using stacked AEs. In this dissertation, I have contributed to deep structures through learning models to deliver more precise predictions. The models introduced in this dissertation are called cascaded and partially cascaded methods of training, which benefit from the fusion of low- and high-level representations. These models for training deep structures surpass the precision accuracy of the standard (typical) method of training stacked autoencoders.
Part 1 of this dissertation discusses the deep regression models. Therefore, vehicular traffic flow prediction (time series data) with respect to spatio-temporal properties of traffic data in highly correlated terrestrial roads and highways is explored for the regression analysis. Abnormalities of traffic data and the correlation between the traffic flow rate and other traffic variables are considered in this research, which makes it different from previous works conducted on traffic flow prediction.
Part 2, generalizes the results of the proposed architectures to the classification task. Therefore, an application using neuroimaging data (nominal data) is considered in the second part of this dissertation. The neuroimaging data under study is magnetic resonance images of the human brain suffering from a common type of dementia, Alzheimer’s disease. The diagnosis of early stages of Alzheimer’s disease plays a key role in patients’ lives, hence the goal of this application. The results reveal remarkably precise predictions compared to the previous studies. Once again partially cascaded models surpass the typical training of stacked AEs by achieving high accuracies in constructing four-class classifiers to diagnose early onset Alzheimer’s disease.
|Commitee:||Akopian, David, Lin, Wei-Ming, Wijemanne, Subhashie|
|School:||The University of Texas at San Antonio|
|Department:||Electrical & Computer Engineering|
|School Location:||United States -- Texas|
|Source:||DAI-B 78/08(E), Dissertation Abstracts International|
|Subjects:||Electrical engineering, Artificial intelligence|
|Keywords:||Alzheimer's disease, Cascaded autoencoders, Deep learning, Feature fusion, Stacked autoencoders, Traffic forcasting|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
The supplemental file or files you are about to download were provided to ProQuest by the author as part of a
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
Copyright of the original materials contained in the supplemental file is retained by the author and your access to the
supplemental files is subject to the ProQuest Terms and Conditions of use.
Depending on the size of the file(s) you are downloading, the system may take some time to download them. Please be