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

Stochastic approaches for correlatio...

Chen, Zhe.

Linked to FindBook

Google Book

Amazon

博客來

Stochastic approaches for correlation-based learning.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Stochastic approaches for correlation-based learning./
Author:	Chen, Zhe.
Description:	170 p.
Notes:	Source: Dissertation Abstracts International, Volume: 66-06, Section: B, page: 3302.
Contained By:	Dissertation Abstracts International66-06B.
Subject:	Engineering, Electronics and Electrical. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=NR04227
ISBN:	0494042273

Stochastic approaches for correlation-based learning.
Chen, Zhe.

Stochastic approaches for correlation-based learning. - 170 p.

Source: Dissertation Abstracts International, Volume: 66-06, Section: B, page: 3302.

Thesis (Ph.D.)--McMaster University (Canada), 2005.

The goal of this thesis is to exploit stochastic (noise-driven or Monte Carlo) approaches for generic derivative-free optimization, and to apply these methods to various machine learning problems, including unsupervised learning for perceptual systems and supervised learning for training neural networks.

ISBN: 0494042273Subjects--Topical Terms:

626636
Engineering, Electronics and Electrical.

Stochastic approaches for correlation-based learning.
LDR:03195nmm 2200277 4500 001 1813182
005 20060427133136.5
008 130610s2005 eng d
020 $a 0494042273
035 $a (UnM)AAINR04227
035 $a AAINR04227
040 $a UnM $c UnM
100 1 $a Chen, Zhe. $3 814621
245 1 0 $a Stochastic approaches for correlation-based learning.
300 $a 170 p.
500 $a Source: Dissertation Abstracts International, Volume: 66-06, Section: B, page: 3302.
502 $a Thesis (Ph.D.)--McMaster University (Canada), 2005.
520 $a The goal of this thesis is to exploit stochastic (noise-driven or Monte Carlo) approaches for generic derivative-free optimization, and to apply these methods to various machine learning problems, including unsupervised learning for perceptual systems and supervised learning for training neural networks.
520 $a We review the root of correlation-based learning and introduce a correlation-based gradient-free optimization procedure, which is known as ALOPEX (ALgorithm Of Pattern EXtraction). As a generic optimization framework, the ALOPEX-type algorithms have certain advantageous features: (i) gradient-free; (ii) network architecture independence; (iii) synchronous learning; and (iv) using noise to help escape local minima or maxima. These appealing features make the ALOPEX be a useful tool for many non-convex optimization and machine learning problems. We have successfully applied the ALOPEX-type algorithms for many perceptual learning tasks. We have, for the first time, applied the algorithm to several classic figure-ground segregation perceptual tasks in sensory perception and reported some novel findings. We also pioneer to apply the ALOPEX algorithm to learn a Neurocompensator for hearing compensation as an ingredient of the hearing-aid design.
520 $a We have provided a systematical overview of Bayesian estimation, Monte Carlo sampling and optimization. In particular, we have applied sequential Monte Carlo sampling methods, within the Bayesian framework, to both state and parameter estimation problems. In sequential state estimation, we have applied particle filtering, with several proposed improvement schemes, to the tracking problems, including a real-life multiple-input-multiple-output (MIMO) wireless channel estimation problem. In parameter estimation, we have proposed two novel Monte Carlo sampling-based ALOPEX algorithms for optimization and training neural networks. Experiments on various learning tasks, including pattern recognition, on-line financial data prediction, on-line system identification, and chaotic time series prediction, have demonstrated the efficacy and strengths of our proposed algorithms.
520 $a In summary, we have addressed a unified philosophical and technical theme in this thesis; we have presented some new theoretical propositions, several novel algorithmic developments, as well as many successful (including some novel) applications.
590 $a School code: 0197.
650 4 $a Engineering, Electronics and Electrical. $3 626636
690 $a 0544
710 2 0 $a McMaster University (Canada). $3 1024893
773 0 $t Dissertation Abstracts International $g 66-06B.
790 $a 0197
791 $a Ph.D.
792 $a 2005
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=NR04227