東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Unsupervised Data-Driven Algorithms ...

Guan, Jiahui.

Linked to FindBook

Google Book

Amazon

博客來

Unsupervised Data-Driven Algorithms with Adaptive Metric for Pattern Recognition, Hypothesis Testing, and Image Segmentation.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Unsupervised Data-Driven Algorithms with Adaptive Metric for Pattern Recognition, Hypothesis Testing, and Image Segmentation./
Author:	Guan, Jiahui.
Published:	Ann Arbor : ProQuest Dissertations & Theses, : 2018,
Description:	147 p.
Notes:	Source: Dissertations Abstracts International, Volume: 80-02, Section: B.
Contained By:	Dissertations Abstracts International80-02B.
Subject:	Biostatistics. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10791922
ISBN:	9780438289956

Unsupervised Data-Driven Algorithms with Adaptive Metric for Pattern Recognition, Hypothesis Testing, and Image Segmentation.
Guan, Jiahui.

Unsupervised Data-Driven Algorithms with Adaptive Metric for Pattern Recognition, Hypothesis Testing, and Image Segmentation. - Ann Arbor : ProQuest Dissertations & Theses, 2018 - 147 p.

Source: Dissertations Abstracts International, Volume: 80-02, Section: B.

Thesis (Ph.D.)--University of California, Davis, 2018.

This item is not available from ProQuest Dissertations & Theses.

In the presence of large and complex data, how to explore, aggregate and extract information remains a challenge. Lack of training labels and unknown data structure makes the problem more difficult. This dissertation focuses on developing unsupervised data-driven algorithm with adaptive metric in different application domains. The first part of the dissertation develops a new algorithm, DCG++ that generates a similarity measure that is data-driven and ultrametric. DCG++ uses Markov Chain Random Walk to capture intrinsic geometry of data, scans possible scales, and combines all the information using a simple procedure that is shown to generate an ultrametric. We validate the effectiveness of this similarity measure on synthetic data with complex geometry, on a real-world data set as well as on an image segmentation problem. The experimental results show a significant improvement on performance with the DCG-based ultrametric compared to using an empirical distance measure. The second part incorporates DCG++ into statistical hypothesis testing. Based on DCG++, we construct coupling geometry on bipartite networks. Such coupling geometry provides multiscale block configurations, which helps hypotheses testing on pattern geometry. A new version block-based nestedness index is proposed and its validity is checked with other state-of-art methods. To study social network and commensal E. coli among rhesus macaques, we use DCG to reconstruct behavioral communities and test E. coli similarity within- versus between- social network groups. Results support the hypothesis that social network communities may act as bottlenecks to contain the spread of infectious agents. In the domain of medical imaging where labels of training data are usually limited, we propose a new unsupervised algorithm for pulmonary nodules detection on three-dimensional computed tomography (CT). The method is segmentation-based detection that adapts statistically region-growing, uses random walk smoothing, and 3D coordinates examination. We benchmark this algorithm on real patients' data from lung image database consortium (LIDC). Results show that the proposed algorithm can successfully detect and segment nodules with arbitrary shape and locations. On the assessment of feature extraction, we develop deep residual autoencoder that contains "bottle-neck" concatenate-identity mapping. Its performance is tested on PTZ zebrafish brain video and experimental results reveal that though capturing major trends, autoencoder may lose certain small features.

ISBN: 9780438289956Subjects--Topical Terms:

1002712
Biostatistics.

Unsupervised Data-Driven Algorithms with Adaptive Metric for Pattern Recognition, Hypothesis Testing, and Image Segmentation.
LDR:03785nmm a2200337 4500 001 2209125
005 20191025102838.5
008 201008s2018 ||||||||||||||||| ||eng d
020 $a 9780438289956
035 $a (MiAaPQ)AAI10791922
035 $a (MiAaPQ)ucdavis:17823
035 $a AAI10791922
040 $a MiAaPQ $c MiAaPQ
100 1 $a Guan, Jiahui. $3 3436205
245 1 0 $a Unsupervised Data-Driven Algorithms with Adaptive Metric for Pattern Recognition, Hypothesis Testing, and Image Segmentation.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2018
300 $a 147 p.
500 $a Source: Dissertations Abstracts International, Volume: 80-02, Section: B.
500 $a Publisher info.: Dissertation/Thesis.
500 $a Advisor: Hsieh, Fushing;Koehl, Patrice.
502 $a Thesis (Ph.D.)--University of California, Davis, 2018.
506 $a This item is not available from ProQuest Dissertations & Theses.
506 $a This item must not be sold to any third party vendors.
520 $a In the presence of large and complex data, how to explore, aggregate and extract information remains a challenge. Lack of training labels and unknown data structure makes the problem more difficult. This dissertation focuses on developing unsupervised data-driven algorithm with adaptive metric in different application domains. The first part of the dissertation develops a new algorithm, DCG++ that generates a similarity measure that is data-driven and ultrametric. DCG++ uses Markov Chain Random Walk to capture intrinsic geometry of data, scans possible scales, and combines all the information using a simple procedure that is shown to generate an ultrametric. We validate the effectiveness of this similarity measure on synthetic data with complex geometry, on a real-world data set as well as on an image segmentation problem. The experimental results show a significant improvement on performance with the DCG-based ultrametric compared to using an empirical distance measure. The second part incorporates DCG++ into statistical hypothesis testing. Based on DCG++, we construct coupling geometry on bipartite networks. Such coupling geometry provides multiscale block configurations, which helps hypotheses testing on pattern geometry. A new version block-based nestedness index is proposed and its validity is checked with other state-of-art methods. To study social network and commensal E. coli among rhesus macaques, we use DCG to reconstruct behavioral communities and test E. coli similarity within- versus between- social network groups. Results support the hypothesis that social network communities may act as bottlenecks to contain the spread of infectious agents. In the domain of medical imaging where labels of training data are usually limited, we propose a new unsupervised algorithm for pulmonary nodules detection on three-dimensional computed tomography (CT). The method is segmentation-based detection that adapts statistically region-growing, uses random walk smoothing, and 3D coordinates examination. We benchmark this algorithm on real patients' data from lung image database consortium (LIDC). Results show that the proposed algorithm can successfully detect and segment nodules with arbitrary shape and locations. On the assessment of feature extraction, we develop deep residual autoencoder that contains "bottle-neck" concatenate-identity mapping. Its performance is tested on PTZ zebrafish brain video and experimental results reveal that though capturing major trends, autoencoder may lose certain small features.
590 $a School code: 0029.
650 4 $a Biostatistics. $3 1002712
650 4 $a Statistics. $3 517247
690 $a 0308
690 $a 0463
710 2 $a University of California, Davis. $b Biostatistics. $3 3434378
773 0 $t Dissertations Abstracts International $g 80-02B.
790 $a 0029
791 $a Ph.D.
792 $a 2018
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10791922