國立高雄大學圖資館 |

Language: English

Back

Self-organizing algorithms for contr...

Choi, Jongeun.

Self-organizing algorithms for controlling network feature map probability distributions, and synthesizing multiple robust controllers.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Self-organizing algorithms for controlling network feature map probability distributions, and synthesizing multiple robust controllers.
Author:	Choi, Jongeun.
Description:	150 p.
Notes:	Adviser: Roberto Horowitz.
Notes:	Source: Dissertation Abstracts International, Volume: 67-08, Section: B, page: 4657.
Contained By:	Dissertation Abstracts International67-08B.
Subject:	Engineering, Electronics and Electrical.
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3228291
ISBN:	9780542822537

Self-organizing algorithms for controlling network feature map probability distributions, and synthesizing multiple robust controllers.
Choi, Jongeun.

Self-organizing algorithms for controlling network feature map probability distributions, and synthesizing multiple robust controllers. - 150 p.

Adviser: Roberto Horowitz.

Thesis (Ph.D.)--University of California, Berkeley, 2006.

Surprisingly, these two apparently different topics, learning algorithms for controlling network feature map probability distributions and the synthesis of a partition of an uncertain set and multiple robust controllers, share a common idea: partitioning an uncertain space and designing local service providers. The main theme in this dissertation is inspired by the process of designing a vector quantizer. The proposed competitive learning algorithms are similar to scalar adaptive quantizers and their encoders are also subject to the nearest neighbor condition. However, the proposed learning algorithms find different optimal decoders that control the network feature map probability distribution according to the preassigned target feature map probability distribution. In the case of multiple robust controllers synthesis, the encoder design is replaced by partitioning the compact parametrically uncertain set and the decoder design is replaced by the optimization of a robust controller for each local subset. The usual quantization error performance cost is replaced by the worst-case H2 and Hinfinity performance costs of the closed-loop system with respect to any parameter variations within the uncertainty.

ISBN: 9780542822537Subjects--Topical Terms:

226981
Engineering, Electronics and Electrical.

Self-organizing algorithms for controlling network feature map probability distributions, and synthesizing multiple robust controllers.
LDR:04904nmm _2200313 _450 001 180641
005 20080111103803.5
008 090528s2006 eng d
020 $a 9780542822537
035 $a 00311666
040 $a UMI $c UMI
100 0 $a Choi, Jongeun. $3 264218
245 1 0 $a Self-organizing algorithms for controlling network feature map probability distributions, and synthesizing multiple robust controllers.
300 $a 150 p.
500 $a Adviser: Roberto Horowitz.
500 $a Source: Dissertation Abstracts International, Volume: 67-08, Section: B, page: 4657.
502 $a Thesis (Ph.D.)--University of California, Berkeley, 2006.
520 # $a Surprisingly, these two apparently different topics, learning algorithms for controlling network feature map probability distributions and the synthesis of a partition of an uncertain set and multiple robust controllers, share a common idea: partitioning an uncertain space and designing local service providers. The main theme in this dissertation is inspired by the process of designing a vector quantizer. The proposed competitive learning algorithms are similar to scalar adaptive quantizers and their encoders are also subject to the nearest neighbor condition. However, the proposed learning algorithms find different optimal decoders that control the network feature map probability distribution according to the preassigned target feature map probability distribution. In the case of multiple robust controllers synthesis, the encoder design is replaced by partitioning the compact parametrically uncertain set and the decoder design is replaced by the optimization of a robust controller for each local subset. The usual quantization error performance cost is replaced by the worst-case H2 and Hinfinity performance costs of the closed-loop system with respect to any parameter variations within the uncertainty.
520 # $a The first problem is to design a parametric diffeomorphism learning algorithm for one-dimensional topology preserving neural networks. Taking the set of output weights of the neural network as a decoder, the nearest neighbor rule is used to partition the input space in which an unknown input signal probability distribution lies. The neural network's output weights converge to a set that produces a predefined winning coordinate probability distribution when the probability density function of the input signal is unknown and not necessarily uniform. In turn the learning process provides an orientation preserving diffeomorphic function from the known neural coordinate domain to the unknown input signal space. This diffeomorphism maps the predefined neural coordinate probability density function into the unknown probability density function of the input signal. The convergence properties of the proposed learning algorithm are analyzed using the ODE approach and verified by a simulation study.
520 # $a The second problem is to simultaneously determine a partition of a parametrically uncertain set and synthesize its corresponding set of multiple robust controllers that optimize the worst-case performance of a linear time invariant system under such parametric uncertainty. The parametric uncertainty region is assumed to be convex polytopic, and it is partitioned into a set of convex polytopic local regions. It is desired that all plants which belong to a local region are to be controlled by a single controller, which is designed to give an optimal worst-case performance for that region. Even though the formulated problem with respect to a certain performance cost is nonconvex, and thus it is difficult to ensure global optimality, descent algorithms are provided to simultaneously update the local regions and the multiple controllers so that they guarantee monotonic non-increasing total performance. For instance for off-line estimation, a resulting local region that the real parameters lie will be detected, and the controller optimized for that region will be selected and applied to the real plant to improve performance. Extensions from LTI parametric uncertain systems to linear parameter-varying (LPV) systems are discussed.
520 # $a This dissertation deals with the problem of partitioning a relatively large uncertain space into smaller subspaces, each of which is serviced by a single agent or service provider, such that the set of multiple service providers functions in a predefined optimal manner. This basic idea is implemented for two different problems.
590 $a School code: 0028.
650 # 0 $a Engineering, Electronics and Electrical. $3 226981
650 # 0 $a Engineering, Mechanical. $3 212470
650 # 0 $a Computer Science. $3 212513
690 $a 0544
690 $a 0548
690 $a 0984
710 0 # $a University of California, Berkeley. $3 212474
773 0 # $g 67-08B. $t Dissertation Abstracts International
790 $a 0028
790 1 0 $a Horowitz, Roberto, $e advisor
791 $a Ph.D.
792 $a 2006
856 4 0 $u http://libsw.nuk.edu.tw:81/login?url=http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3228291 $z http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3228291