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

Machine Learning Methods for Uncerta...

Li, Rui.

Linked to FindBook

Google Book

Amazon

博客來

Machine Learning Methods for Uncertainty Estimation and Decision-Making.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Machine Learning Methods for Uncertainty Estimation and Decision-Making./
Author:	Li, Rui.
Published:	Ann Arbor : ProQuest Dissertations & Theses, : 2019,
Description:	103 p.
Notes:	Source: Dissertations Abstracts International, Volume: 81-03, Section: B.
Contained By:	Dissertations Abstracts International81-03B.
Subject:	Statistics. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=27529146
ISBN:	9781085646147

Machine Learning Methods for Uncertainty Estimation and Decision-Making.
Li, Rui.

Machine Learning Methods for Uncertainty Estimation and Decision-Making. - Ann Arbor : ProQuest Dissertations & Theses, 2019 - 103 p.

Source: Dissertations Abstracts International, Volume: 81-03, Section: B.

Thesis (Ph.D.)--North Carolina State University, 2019.

This item must not be sold to any third party vendors.

Machine learning has become widely used in many applications, due to its model flexibility and robustness. In this thesis, we study and develop three different machine learning methods that can be used in uncertainty estimation and decision-making. In Chapter 1, we develop a nonparametric variable selection procedure to select for variables that impact clinical treatment decisions. Variable selection research has largely focused on selecting variables that are important for prediction, and less attention has been paid to identifying variables that are important for decision making. Our approach is based on a hypothesis testing view applied to the regret function of different models. We demonstrate the use of this approach as an example using Gaussian process regression together with backward elimination. The performance of our method is evaluated in simulation studies and an application with AIDS Clinical Trial Group Study. In Chapter 2, we develop a model agnostic framework to estimate the conditional distribution of a response variable given a set of predictor variables. Our approach transforms a conditional distribution estimation problem into a constrained multi-class classification problem, in which tools such as deep neural networks can be utilized.We propose a novel joint binary cross-entropy loss function to accomplish this goal. We demonstrate its performance in various simulation studies comparing to state-of-the-art competing methods. Additionally, our method shows improved accuracy in a probabilistic solar energy forecasting problem.In Chapter 3, we utilize the function approximation property of neural network models to directly estimate the conditional cumulative distribution function, and thus achieve accurate conditional distribution estimation.We propose novel neural network structures and monotonic constraints to ensure the estimated distribution function is valid.We also reduce computation burden by an adaptive training algorithm.We evaluate the proposed method and show superior performance compared to other flexible models for conditional distribution estimation. We also demonstrate the usefulness of our model in the probabilistic forecasting of electricity load demand.

ISBN: 9781085646147Subjects--Topical Terms:

517247
Statistics.
Subjects--Index Terms:

Machine learning

Machine Learning Methods for Uncertainty Estimation and Decision-Making.
LDR:03405nmm a2200337 4500 001 2268767
005 20200824072448.5
008 220629s2019 ||||||||||||||||| ||eng d
020 $a 9781085646147
035 $a (MiAaPQ)AAI27529146
035 $a (MiAaPQ)NCState_Univ18402036738
035 $a AAI27529146
040 $a MiAaPQ $c MiAaPQ
100 1 $a Li, Rui. $3 1023852
245 1 0 $a Machine Learning Methods for Uncertainty Estimation and Decision-Making.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2019
300 $a 103 p.
500 $a Source: Dissertations Abstracts International, Volume: 81-03, Section: B.
500 $a Advisor: Flores, Kevin;Maity, Arnab;Lu, Wenbin;Reich, Brian;Bondell, Howard.
502 $a Thesis (Ph.D.)--North Carolina State University, 2019.
506 $a This item must not be sold to any third party vendors.
520 $a Machine learning has become widely used in many applications, due to its model flexibility and robustness. In this thesis, we study and develop three different machine learning methods that can be used in uncertainty estimation and decision-making. In Chapter 1, we develop a nonparametric variable selection procedure to select for variables that impact clinical treatment decisions. Variable selection research has largely focused on selecting variables that are important for prediction, and less attention has been paid to identifying variables that are important for decision making. Our approach is based on a hypothesis testing view applied to the regret function of different models. We demonstrate the use of this approach as an example using Gaussian process regression together with backward elimination. The performance of our method is evaluated in simulation studies and an application with AIDS Clinical Trial Group Study. In Chapter 2, we develop a model agnostic framework to estimate the conditional distribution of a response variable given a set of predictor variables. Our approach transforms a conditional distribution estimation problem into a constrained multi-class classification problem, in which tools such as deep neural networks can be utilized.We propose a novel joint binary cross-entropy loss function to accomplish this goal. We demonstrate its performance in various simulation studies comparing to state-of-the-art competing methods. Additionally, our method shows improved accuracy in a probabilistic solar energy forecasting problem.In Chapter 3, we utilize the function approximation property of neural network models to directly estimate the conditional cumulative distribution function, and thus achieve accurate conditional distribution estimation.We propose novel neural network structures and monotonic constraints to ensure the estimated distribution function is valid.We also reduce computation burden by an adaptive training algorithm.We evaluate the proposed method and show superior performance compared to other flexible models for conditional distribution estimation. We also demonstrate the usefulness of our model in the probabilistic forecasting of electricity load demand.
590 $a School code: 0155.
650 4 $a Statistics. $3 517247
653 $a Machine learning
653 $a Clinical treatment decisions
653 $a Electricity load demand
690 $a 0463
710 2 $a North Carolina State University. $3 1018772
773 0 $t Dissertations Abstracts International $g 81-03B.
790 $a 0155
791 $a Ph.D.
792 $a 2019
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=27529146