東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

Theoretical and experimental analysi...

Andonova Jaeger, Savina Kirilova.

FindBook

Google Book

Amazon

博客來

Theoretical and experimental analysis of the generalization ability of some statistical learning algorithms.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Theoretical and experimental analysis of the generalization ability of some statistical learning algorithms./
作者:	Andonova Jaeger, Savina Kirilova.
面頁冊數:	261 p.
附註:	Source: Dissertation Abstracts International, Volume: 65-01, Section: B, page: 0275.
Contained By:	Dissertation Abstracts International65-01B.
標題:	Statistics. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3118799

Theoretical and experimental analysis of the generalization ability of some statistical learning algorithms.
Andonova Jaeger, Savina Kirilova.

Theoretical and experimental analysis of the generalization ability of some statistical learning algorithms. - 261 p.

Source: Dissertation Abstracts International, Volume: 65-01, Section: B, page: 0275.

Thesis (Ph.D.)--Boston University, 2004.

This thesis considers problems arising in the area of non-parametric regression and classification when using combinations of learning machines.Subjects--Topical Terms:

517247
Statistics.

Theoretical and experimental analysis of the generalization ability of some statistical learning algorithms.
LDR:03005nmm 2200325 4500 001 1861501
005 20041117064927.5
008 130614s2004 eng d
035 $a (UnM)AAI3118799
035 $a AAI3118799
040 $a UnM $c UnM
100 1 $a Andonova Jaeger, Savina Kirilova. $3 1949096
245 1 0 $a Theoretical and experimental analysis of the generalization ability of some statistical learning algorithms.
300 $a 261 p.
500 $a Source: Dissertation Abstracts International, Volume: 65-01, Section: B, page: 0275.
500 $a Major Professor: Mark Kon.
502 $a Thesis (Ph.D.)--Boston University, 2004.
520 $a This thesis considers problems arising in the area of non-parametric regression and classification when using combinations of learning machines.
520 $a New measures of complexities of convex combinations of classifiers from Generalization bounds taking into account these complexities are derived. The approaches taken here for the exploration of these problems are based upon statistical learning theory, empirical process theory, approximation theory and regularization theory. Previous results are described and compared theoretically and experimentally with new results presented here. The generalization bounds derived and these new measures of complexity of convex combinations of classifiers are shown to be useful for model selection.
520 $a Classifiers or regressors can be combined using component-based hierarchies of kernel machines. Interest in applying hierarchies in learning problems has grown recently. Some new theoretical results on the generalization performance of component-based hierarchies are presented here. The generalization bounds derived here depend on the empirical margin and the sum of the weights of the component kernel classifiers derived at different levels of the hierarchy.
520 $a Lastly, generalization bounds are shown for convex combinations of classifiers from random classes, i.e. function classes depending on the training set and invariant under permutation of the set and possessing the increasing property. Examples of convex combinations of classifiers from random classes are classifiers generated by Support Vector Machines, boosting with discretized decision stumps, radial basis function networks or some hierarchies of kernel machines. The approach taken for deriving the generalization bounds introduces the equivalent to Vapnik's "optimistic" inequality for random classes of sets. Some measures of complexities, such as sparsity of the weights and cluster variance for the convex combinations of classifiers, are considered as well. Whenever possible, these bounds are compared with previous bounds.
590 $a School code: 0017.
650 4 $a Statistics. $3 517247
650 4 $a Applied Mechanics. $3 1018410
650 4 $a Computer Science. $3 626642
650 4 $a Mathematics. $3 515831
690 $a 0463
690 $a 0346
690 $a 0984
690 $a 0405
710 2 0 $a Boston University. $3 1017454
773 0 $t Dissertation Abstracts International $g 65-01B.
790 1 0 $a Kon, Mark, $e advisor
790 $a 0017
791 $a Ph.D.
792 $a 2004
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3118799