國立高雄大學圖資館 |

登入

回首頁

Stochastic search for optimum under a control objective

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Stochastic search for optimum under a control objective
作者:	Cope, Eric William.
面頁冊數:	175 p.
附註:	Adviser: Nicholas Bambos.
附註:	Source: Dissertation Abstracts International, Volume: 65-04, Section: B, page: 2073.
Contained By:	Dissertation Abstracts International65-04B.
標題:	Operations Research.
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3128638
ISBN:	0496759000

Stochastic search for optimum under a control objective
Cope, Eric William.

Stochastic search for optimum under a control objective [electronic resource] - 175 p.

Adviser: Nicholas Bambos.

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

In Chapter 3, the algorithms developed for the one-dimensional control problem are extended to a multidimensional search setting involving Markov decision processes. This algorithm performs a direct search in the space of stationary, deterministic control policies. An analysis shows that the policies enacted by this algorithm converge with probability one to a set of stationary points of an average reward measure over the policy space.

ISBN: 0496759000Subjects--Topical Terms:

227148
Operations Research.

Stochastic search for optimum under a control objective
LDR:03238nmm _2200313 _450 001 162632
005 20051017073506.5
008 230606s2004 eng d
020 $a 0496759000
035 $a 00149133
035 $a 162632
040 $a UnM $c UnM
100 0 $a Cope, Eric William. $3 227776
245 1 0 $a Stochastic search for optimum under a control objective $h [electronic resource]
300 $a 175 p.
500 $a Adviser: Nicholas Bambos.
500 $a Source: Dissertation Abstracts International, Volume: 65-04, Section: B, page: 2073.
502 $a Thesis (Ph.D.)--Stanford University, 2004.
520 # $a In Chapter 3, the algorithms developed for the one-dimensional control problem are extended to a multidimensional search setting involving Markov decision processes. This algorithm performs a direct search in the space of stationary, deterministic control policies. An analysis shows that the policies enacted by this algorithm converge with probability one to a set of stationary points of an average reward measure over the policy space.
520 # $a In Chapter 4, we investigate the sequential pricing of information goods on the Internet when demand for those goods is unknown. The emphasis is on identifying algorithms that achieve good small-sample performance. A nonparametric Bayesian model is introduced to capture the demand uncertainty, and several heuristic search methods based on "index functions" are proposed and compared. In general, we find that these search methods are efficient and robust.
520 # $a In certain problems of "black box" control, an agent attempts to maximize a random stream of rewards that depend, in an initially unknown way, upon a sequence of actions taken by the agent. This dissertation treats situations where the control problem may be understood as a search for an optimal control policy within a compact set of policies. Because real rewards are accrued during the course of the search, an efficient search is one where the chosen sequence of controls converges as rapidly possible to an optimal control policy.
520 # $a This dissertation is divided into three major chapters which follow an introductory chapter which introduces the main issues. Chapter 2 addresses a simple problem where the search takes place in a one-dimensional control space, and the expected rewards may be assumed to be unimodal over this space. A class of lower bounds for the asymptotic growth rate of a "cost" performance measure is derived for any control strategy. It is shown that a version of the Kiefer-Wolfowitz method of stochastic approximation achieves one such lower bound. Classes of control strategies are then introduced that nearly achieve the lower bounds for the cost growth rate more generally. These results show that the lower bounds are tight and that the control strategies are asymptotically optimal.
590 $a School code: 0212.
650 # 0 $a Operations Research. $3 227148
650 # 0 $a Statistics. $3 182057
690 $a 0463
690 $a 0796
710 0 # $a Stanford University. $3 212607
773 0 # $g 65-04B. $t Dissertation Abstracts International
790 $a 0212
790 1 0 $a Bambos, Nicholas, $e advisor
791 $a Ph.D.
792 $a 2004
856 4 0 $u http://libsw.nuk.edu.tw/login?url=http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3128638 $z http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3128638