國立高雄大學圖資館 |

Language: English

Back

Diagnostics and extrapolation in mac...

Hooker, Giles.

Diagnostics and extrapolation in machine learning.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Diagnostics and extrapolation in machine learning.
Author:	Hooker, Giles.
Description:	127 p.
Notes:	Adviser: Jerome Friedman.
Notes:	Source: Dissertation Abstracts International, Volume: 65-09, Section: B, page: 4648.
Contained By:	Dissertation Abstracts International65-09B.
Subject:	Statistics.
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3145521
ISBN:	0496044648

Diagnostics and extrapolation in machine learning.
Hooker, Giles.

Diagnostics and extrapolation in machine learning. - 127 p.

Adviser: Jerome Friedman.

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

All the ideas in this work are designed to be fully general and compatible with any machine learning algorithm.

ISBN: 0496044648Subjects--Topical Terms:

182057
Statistics.

Diagnostics and extrapolation in machine learning.
LDR:02278nmm _2200313 _450 001 162875
005 20051017073532.5
008 090528s2004 eng d
020 $a 0496044648
035 $a 00149376
040 $a UnM $c UnM
100 0 $a Hooker, Giles. $3 228020
245 1 0 $a Diagnostics and extrapolation in machine learning.
300 $a 127 p.
500 $a Adviser: Jerome Friedman.
500 $a Source: Dissertation Abstracts International, Volume: 65-09, Section: B, page: 4648.
502 $a Thesis (Ph.D.)--Stanford University, 2004.
520 # $a All the ideas in this work are designed to be fully general and compatible with any machine learning algorithm.
520 # $a Finally, I advocate a modification to the Functional ANOVA that uses this estimate to avoid the effects of bad extrapolation while retaining many of the useful properties of the decomposition.
520 # $a I present a suite of tools for understanding high dimensional prediction functions that are based on the Functional ANOVA decomposition and argue that these are optimal in an idealized setting. I then show that they can be distorted to an arbitrary extent if the predictor space contains large regions of extrapolation.
520 # $a The subject of this thesis is the interaction between the problems of diagnostics and extrapolation in Machine Learning.
520 # $a This thesis gives a criterion of extrapolation and details tree-based methods to evaluate it. This methodology provides a comprehensible representation of the distribution of training data and a diagnostic for functional behavior in regions of low data density. I then discuss the issue of making predictions at points of extrapolation. I suggest a strategy for stabilizing a general learning algorithm away from training data that is motivated by a Bayesian heuristic not unlike ridge regression and which bears some resemblance to Kriging.
590 $a School code: 0212.
650 # 0 $a Statistics. $3 182057
650 # 0 $a Computer Science. $3 212513
690 $a 0463
690 $a 0984
710 0 # $a Stanford University. $3 212607
773 0 # $g 65-09B. $t Dissertation Abstracts International
790 $a 0212
790 1 0 $a Friedman, Jerome, $e advisor
791 $a Ph.D.
792 $a 2004
856 4 0 $u http://libsw.nuk.edu.tw:81/login?url=http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3145521 $z http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3145521