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

Multivariate kernel methods in the a...

Purdom, Elizabeth.

Linked to FindBook

Google Book

Amazon

博客來

Multivariate kernel methods in the analysis of graphical structures.

Record Type:	Language materials, printed : Monograph/item
Title/Author:	Multivariate kernel methods in the analysis of graphical structures./
Author:	Purdom, Elizabeth.
Description:	209 p.
Notes:	Adviser: Susan Holmes.
Contained By:	Dissertation Abstracts International67-09B.
Subject:	Statistics. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3235325
ISBN:	9780542895401

Multivariate kernel methods in the analysis of graphical structures.
Purdom, Elizabeth.

Multivariate kernel methods in the analysis of graphical structures. - 209 p.

Adviser: Susan Holmes.

Thesis (Ph.D.)--Stanford University, 2006.

Increasingly, current knowledge of a biological process is summarized in the form of a graph or a network. However, graphical structures are difficult to incorporate into a standard data analysis. In particular, it is difficult to relate the information in a graph to new experimental data. Various authors have suggested kernel approaches to these problems. Such kernel methods adapt classical multivariate statistical techniques for more general types of data. Rather than require numerical description of the data, kernel techniques require only a particular type of measure of similarity between objects. For biological graphs, different authors have used a measure of similarity between nodes of a graph to analyze graphs with kernel techniques. However, the kernel methods in use only generally describe the relationship between the graph and the experiment without being explicitly predictive; in particular, they derive from the classical technique of Canonical Correlation Analysis (CCA).

ISBN: 9780542895401Subjects--Topical Terms:

517247
Statistics.

Multivariate kernel methods in the analysis of graphical structures.
LDR:03143nam 2200289 a 45 001 968880
005 20110920
008 110921s2006 eng d
020 $a 9780542895401
035 $a (UnM)AAI3235325
035 $a AAI3235325
040 $a UnM $c UnM
100 1 $a Purdom, Elizabeth. $3 1292935
245 1 0 $a Multivariate kernel methods in the analysis of graphical structures.
300 $a 209 p.
500 $a Adviser: Susan Holmes.
500 $a Source: Dissertation Abstracts International, Volume: 67-09, Section: B, page: 5174.
502 $a Thesis (Ph.D.)--Stanford University, 2006.
520 $a Increasingly, current knowledge of a biological process is summarized in the form of a graph or a network. However, graphical structures are difficult to incorporate into a standard data analysis. In particular, it is difficult to relate the information in a graph to new experimental data. Various authors have suggested kernel approaches to these problems. Such kernel methods adapt classical multivariate statistical techniques for more general types of data. Rather than require numerical description of the data, kernel techniques require only a particular type of measure of similarity between objects. For biological graphs, different authors have used a measure of similarity between nodes of a graph to analyze graphs with kernel techniques. However, the kernel methods in use only generally describe the relationship between the graph and the experiment without being explicitly predictive; in particular, they derive from the classical technique of Canonical Correlation Analysis (CCA).
520 $a In this work, we turn instead to explicitly predictive multivariate regression techniques. We extend multivariate regression to the kernel framework and show that this provides a method more appropriate for these tasks, which are generally predictive in nature. In doing so, we also show how the CCA and regression framework are related through regularization of the problem. We illustrate how these kernel techniques are equivalent to classical statistical techniques, if the "data" for an observation consists of the similarity of the observation to the training set. This is a different interpretation of the kernel techniques and can be more convenient for those already familiar with classical statistics. We demonstrate our techniques for two different types of graphs: (1) a yeast metabolic network and (2) a phylogenetic tree. For the metabolic network, we use a network and related experimental data that has been previously analyzed the literature using kernel methods. For the phylogenetic tree, we analyze ecological data collected on bacteria and their estimated phylogenetic tree. For the latter set of data, we suggest a different type of similarity appropriate for phylogenetic trees, and we connect this type of analysis to others in the literature.
590 $a School code: 0212.
650 4 $a Statistics. $3 517247
650 4 $a Biology, Bioinformatics. $3 1018415
690 $a 0463
690 $a 0715
710 2 0 $a Stanford University. $3 754827
773 0 $t Dissertation Abstracts International $g 67-09B.
790 $a 0212
790 1 0 $a Holmes, Susan, $e advisor
791 $a Ph.D.
792 $a 2006
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3235325