東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

Learning Data-Driven Models for Deci...

Virani, Nurali.

FindBook

Google Book

Amazon

博客來

Learning Data-Driven Models for Decision-Making in Intelligent Physical Systems.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Learning Data-Driven Models for Decision-Making in Intelligent Physical Systems./
作者:	Virani, Nurali.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2017,
面頁冊數:	215 p.
附註:	Source: Dissertation Abstracts International, Volume: 78-11(E), Section: B.
Contained By:	Dissertation Abstracts International78-11B(E).
標題:	Artificial intelligence. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10629142
ISBN:	9781369992458

Learning Data-Driven Models for Decision-Making in Intelligent Physical Systems.
Virani, Nurali.

Learning Data-Driven Models for Decision-Making in Intelligent Physical Systems. - Ann Arbor : ProQuest Dissertations & Theses, 2017 - 215 p.

Source: Dissertation Abstracts International, Volume: 78-11(E), Section: B.

Thesis (Ph.D.)--The Pennsylvania State University, 2017.

Intelligent physical systems use machine learning for a variety of tasks from health monitoring to control. As the dependence on autonomous decision-making agents increases, it is of importance to understand and quantify the uncertainty associated with the decisions from machine learning frameworks. In order to facilitate the interaction with human agents (e.g., maintenance engineers and medical doctors) as well as to enable robust control for safety (e.g., autonomous navigation and sensor network adaptation), density estimation enables quantification of uncertainty in the output of a learning framework. In statistical learning, density estimation is a core problem, where the objective is to identify the underlying distribution from which the data are being generated. In this work, density estimation is established as a practical tool for data-driven modeling. A new and simple technique for density estimation is developed using concepts from statistical learning and optimization theory. Along with detection, classification, estimation, and tracking, which are crucial in learning and control, these models can also quantify uncertainty in their outputs.

ISBN: 9781369992458Subjects--Topical Terms:

516317
Artificial intelligence.

Learning Data-Driven Models for Decision-Making in Intelligent Physical Systems.
LDR:05028nmm a2200325 4500 001 2156941
005 20180529081902.5
008 190424s2017 ||||||||||||||||| ||eng d
020 $a 9781369992458
035 $a (MiAaPQ)AAI10629142
035 $a AAI10629142
040 $a MiAaPQ $c MiAaPQ
100 1 $a Virani, Nurali. $3 3344712
245 1 0 $a Learning Data-Driven Models for Decision-Making in Intelligent Physical Systems.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2017
300 $a 215 p.
500 $a Source: Dissertation Abstracts International, Volume: 78-11(E), Section: B.
500 $a Adviser: Asok Ray.
502 $a Thesis (Ph.D.)--The Pennsylvania State University, 2017.
520 $a Intelligent physical systems use machine learning for a variety of tasks from health monitoring to control. As the dependence on autonomous decision-making agents increases, it is of importance to understand and quantify the uncertainty associated with the decisions from machine learning frameworks. In order to facilitate the interaction with human agents (e.g., maintenance engineers and medical doctors) as well as to enable robust control for safety (e.g., autonomous navigation and sensor network adaptation), density estimation enables quantification of uncertainty in the output of a learning framework. In statistical learning, density estimation is a core problem, where the objective is to identify the underlying distribution from which the data are being generated. In this work, density estimation is established as a practical tool for data-driven modeling. A new and simple technique for density estimation is developed using concepts from statistical learning and optimization theory. Along with detection, classification, estimation, and tracking, which are crucial in learning and control, these models can also quantify uncertainty in their outputs.
520 $a This dissertation uses density estimation for developing new methods to solve practical problems of learning and decision-making. A few restrictive assumptions have been eliminated from these problems, yet tractable and accurate methods have been developed in this research. Specifically, in the sequential classification problem, the naive Bayes' assumption of conditional independence between measurements, given state, is relaxed. A novel technique to learn a unified context from multi-modal sensor data is developed. This knowledge of context is used to achieve tractable and accurate multi-modal sensor fusion, which cannot be achieved using the naive Bayes' assumption. Additionally, the context-aware measurement models are also used for unifying state estimation and dynamic sensor selection problems in a stochastic control framework. In sequential hypothesis testing with streaming data, the assumption that the observation sequence is independent and identically distributed (IID) has been removed by developing sequential tests for Markov models of time-series data. Further, density estimation has been used to create Markov models from multidimensional time-series data by developing a unified formulation for alphabet-size selection and measurement-space partitioning. In sequential tracking, the assumption of additive Gaussian noise has been eliminated by learning nonparametric density estimation-based measurement models, which can capture all the uncertainties in a given set of data. These measurement models have been used for state estimation and tracking with particle filters. In a sequential measurement model learning setting, the labels provided by instructors are allowed to be incorrect as the assumption of the instructor being perfect has not been used. A recursive density estimation algorithm has been developed and analyzed to show that correct models can be obtained even with noisy labels.
520 $a In physical systems, the assumptions noted above, usually do not hold due to causal dependencies, physical constraints, operating conditions, various uncertainties, etc., and hence have been selectively relaxed. The theoretical frameworks developed in this research have been validated using simulations and real-world experiments. The practical applications covered in this dissertation include target detection and classification in border surveillance, indoor localization of smart wheelchairs with user-assistance for safety during navigation, and detection of combustion instability using streaming data. These widely differing real-world problems have been used to illustrate the general applicability of the results developed in this thesis. It is envisioned that the formulations and results from this dissertation will be useful in data-driven modeling, real-time decision-making, and robust control of physical systems, making them more intelligent.
590 $a School code: 0176.
650 4 $a Artificial intelligence. $3 516317
650 4 $a Mechanical engineering. $3 649730
650 4 $a Systems science. $3 3168411
690 $a 0800
690 $a 0548
690 $a 0790
710 2 $a The Pennsylvania State University. $b Mechanical Engineering. $3 2103981
773 0 $t Dissertation Abstracts International $g 78-11B(E).
790 $a 0176
791 $a Ph.D.
792 $a 2017
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10629142