語系:
繁體中文
English
說明(常見問題)
登入
回首頁
切換:
標籤
|
MARC模式
|
ISBD
Robust Fragmentation : = A Data-Driv...
~
ProQuest Information and Learning Co.
Robust Fragmentation : = A Data-Driven Approach to Decision-Making under Distributional Ambiguity.
紀錄類型:
書目-語言資料,手稿 : Monograph/item
正題名/作者:
Robust Fragmentation :/
其他題名:
A Data-Driven Approach to Decision-Making under Distributional Ambiguity.
作者:
Moulton, Jeffrey.
面頁冊數:
1 online resource (162 pages)
附註:
Source: Dissertation Abstracts International, Volume: 78-01(E), Section: B.
Contained By:
Dissertation Abstracts International78-01B(E).
標題:
Mathematics. -
電子資源:
click for full text (PQDT)
ISBN:
9781369106527
Robust Fragmentation : = A Data-Driven Approach to Decision-Making under Distributional Ambiguity.
Moulton, Jeffrey.
Robust Fragmentation :
A Data-Driven Approach to Decision-Making under Distributional Ambiguity. - 1 online resource (162 pages)
Source: Dissertation Abstracts International, Volume: 78-01(E), Section: B.
Thesis (Ph.D.)
Includes bibliographical references
Decision makers often must consider many different possible future scenarios when they make a decision. A manager must choose inventory levels to maximize profit when the demand is unknown, investors choose a portfolio to maximize gain in the face of uncertain stock price fluctuations, and service providers must pick facility locations to minimize distances traveled where the location of the next customer is random. When uncertainty is involved, the desired outcome will be affected by factors and variables outside the decider's control and knowledge at the time of the decision. Such factors of uncertainty are hard to account for, especially when exact information about the uncertainty is unknown. It may be difficult to account for all possibilities or to estimate how likely each scenario is. There may be an infinite number of scenarios, too many for the human brain to contemplate. Generally, a precise distribution of these factors is unavailable. The unknown data, which can be quantified as a vector of parameters, follows an unknown distribution leading to the term distributional ambiguity. The decision maker may only have a sample of past events to guide them in balancing the costs and benefits of every possibility. Such limited historical data can help to provide guidance, but how best to use it? One would desire a decision strategy that will efficiently use the data and perform well in terms of the expectation, but also be robust to possible noise, changing conditions, and small sample size. To this end, we develop robust fragmentation (RF), a data-driven approach to decision-making under distributional ambiguity.
Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2018
Mode of access: World Wide Web
ISBN: 9781369106527Subjects--Topical Terms:
527692
Mathematics.
Index Terms--Genre/Form:
554714
Electronic books.
Robust Fragmentation : = A Data-Driven Approach to Decision-Making under Distributional Ambiguity.
LDR
:06397ntm a2200397Ki 4500
001
911028
005
20180517121918.5
006
m o u
007
cr mn||||a|a||
008
190606s2016 xx obm 000 0 eng d
020
$a
9781369106527
035
$a
(MiAaPQ)AAI10155578
035
$a
(MiAaPQ)umn:17412
035
$a
AAI10155578
040
$a
MiAaPQ
$b
eng
$c
MiAaPQ
099
$a
TUL
$f
hyy
$c
available through World Wide Web
100
1
$a
Moulton, Jeffrey.
$3
1182623
245
1 0
$a
Robust Fragmentation :
$b
A Data-Driven Approach to Decision-Making under Distributional Ambiguity.
264
0
$c
2016
300
$a
1 online resource (162 pages)
336
$a
text
$b
txt
$2
rdacontent
337
$a
computer
$b
c
$2
rdamedia
338
$a
online resource
$b
cr
$2
rdacarrier
500
$a
Source: Dissertation Abstracts International, Volume: 78-01(E), Section: B.
500
$a
Adviser: Gilad Lerman.
502
$a
Thesis (Ph.D.)
$c
University of Minnesota
$d
2016.
504
$a
Includes bibliographical references
520
$a
Decision makers often must consider many different possible future scenarios when they make a decision. A manager must choose inventory levels to maximize profit when the demand is unknown, investors choose a portfolio to maximize gain in the face of uncertain stock price fluctuations, and service providers must pick facility locations to minimize distances traveled where the location of the next customer is random. When uncertainty is involved, the desired outcome will be affected by factors and variables outside the decider's control and knowledge at the time of the decision. Such factors of uncertainty are hard to account for, especially when exact information about the uncertainty is unknown. It may be difficult to account for all possibilities or to estimate how likely each scenario is. There may be an infinite number of scenarios, too many for the human brain to contemplate. Generally, a precise distribution of these factors is unavailable. The unknown data, which can be quantified as a vector of parameters, follows an unknown distribution leading to the term distributional ambiguity. The decision maker may only have a sample of past events to guide them in balancing the costs and benefits of every possibility. Such limited historical data can help to provide guidance, but how best to use it? One would desire a decision strategy that will efficiently use the data and perform well in terms of the expectation, but also be robust to possible noise, changing conditions, and small sample size. To this end, we develop robust fragmentation (RF), a data-driven approach to decision-making under distributional ambiguity.
520
$a
We consider a stochastic programming framework, where the decision maker aims to optimize an expected outcome. However, we assume that the governing distribution of the random parameters affecting the outcome is unknown. Only a historical sample of past realizations of such parameters is available. Our goal is to leverage the historical data to effectively and robustly approximate the true problem. This is done by fragmenting the support of the data into pieces to dissect the structure. We reduce the data by replacing it with summary statistics by region. These parameters are used to construct an ambiguity set for the distribution. The ambiguity set consists of all distributions which would satisfy the same regional reduction. We therefore reduce the problem size and avoid overfitting to the training sample. Our approach allows for two types of ambiguity: ambiguity in support and ambiguity in probability. After constructing the ambiguity set, we consider the worst case expectation to provide distributional robustness. Constraining the distribution regionally allows us to capture detailed information about the distribution and keeps the approach from being too conservative. The ambiguity may be tailored to the structure, amount, and reliability of the data.
520
$a
Robust fragmentation is a generalization and extension of several classical and newer approaches to approximating a stochastic program, including the sample average approximation (SAA), moments-based distributionally robust optimization (MDRO), and distributionally robust optimization based on probabilities of individual events. We demonstrate how RF conceptually fits into the greater picture and provides an adaptive way of balancing the benefits and drawbacks of each kind of approach. We make comparisons both theoretically and through numerical experiments.
520
$a
We outline a procedure for breaking up the data and formulating the RF optimization problem for polytopal, ellipsoidal, and other types of fragmentations. We prove convergence results to demonstrate the consistency of our approach. RF can handle overlapping regions in the fragmentation by adapting the way statistics are computed. Our algorithm for fragmenting the data relies heavily on strategic clustering of the data sample to dissect the modality. We discuss methods and heuristics for implementing our approach. We extend and compare different clustering methods in terms of how well they serve as a preliminary step to our procedure. We consider applications in management science and financial mathematics and perform numerical experiments to demonstrate how the approach fits into existing frameworks and to evaluate its performance. It turns out that the new approach has a clear advantage when the data is multimodal and the training sample is small, but not too small relative to the number of modes. Additionally, we illustrate some extensions to the general model. The black swan approach involves adding an extra layer of robustness to account for rare and unexpected events of large magnitude and consequence. Two-stage and multistage stochastic programs extend the framework to decisions that must be made in stages, where after each stage random occurences influence the parameters of the next stage. We derive formulations for these extended models and investigate the production transportation-problem as an application.
533
$a
Electronic reproduction.
$b
Ann Arbor, Mich. :
$c
ProQuest,
$d
2018
538
$a
Mode of access: World Wide Web
650
4
$a
Mathematics.
$3
527692
650
4
$a
Industrial engineering.
$3
679492
650
4
$a
Statistics.
$3
556824
655
7
$a
Electronic books.
$2
local
$3
554714
690
$a
0405
690
$a
0546
690
$a
0463
710
2
$a
ProQuest Information and Learning Co.
$3
1178819
710
2
$a
University of Minnesota.
$b
Mathematics.
$3
1180493
773
0
$t
Dissertation Abstracts International
$g
78-01B(E).
856
4 0
$u
http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10155578
$z
click for full text (PQDT)
筆 0 讀者評論
多媒體
評論
新增評論
分享你的心得
Export
取書館別
處理中
...
變更密碼[密碼必須為2種組合(英文和數字)及長度為10碼以上]
登入