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

Linked to FindBook

Google Book

Amazon

博客來

Information Theory for Trustworthy Machine Learning.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Information Theory for Trustworthy Machine Learning./
Author:	Wang, Hao.
Description:	1 online resource (246 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 83-12, Section: B.
Contained By:	Dissertations Abstracts International83-12B.
Subject:	Applied mathematics. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29210084click for full text (PQDT)
ISBN:	9798819382844

Information Theory for Trustworthy Machine Learning.
Wang, Hao.

Information Theory for Trustworthy Machine Learning. - 1 online resource (246 pages)

Source: Dissertations Abstracts International, Volume: 83-12, Section: B.

Thesis (Ph.D.)--Harvard University, 2022.

Includes bibliographical references

In this thesis, we provide an information-theoretic foundation for trustworthy machine learning (ML). ML algorithms are increasingly used in applications of significant social consequences. It is crucial to ensure that these algorithms are not just accurate but also generalizable, fair, and privacy-preserving. We develop new information-theoretic tools for proving rigorous performance guarantees for practical ML models and establishing protocols to ensure the responsible use of data.Towards rigorous performance guarantees, we derive generalization bounds for understanding the behaviors of complex ML models (e.g., neural networks). In particular, we study how the interaction between data distributions and optimization methods influences generalization. We consider a family of optimization methods---noisy iterative algorithms---and investigate their generalization capability. We derive distribution-dependent generalization bounds by connecting noisy iterative algorithms with additive noise channels found in information theory. Numerical experiments demonstrate that our results can help understand many empirical observations of neural networks. Towards responsible use, we characterize a fundamental limit of algorithmic fairness and privacy. Specifically, we provide conditions that ensure fair use of group attributes and analyze the "best" privacy-utility trade-offs among all privacy-preserving mechanisms. These theoretical results, in turn, inspire new algorithms that can repair unfair models or preserve privacy in data sharing. Finally, we evaluate the robustness of information-theoretic privacy measures and establish statistical consistency of "optimal" privacy mechanisms.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2023

Mode of access: World Wide Web

ISBN: 9798819382844Subjects--Topical Terms:

2122814
Applied mathematics.
Subjects--Index Terms:

Information theoryIndex Terms--Genre/Form:

542853
Electronic books.

Information Theory for Trustworthy Machine Learning.
LDR:03018nmm a2200385K 4500 001 2353200
005 20221214062822.5
006 m o d
007 cr mn ---uuuuu
008 241011s2022 xx obm 000 0 eng d
020 $a 9798819382844
035 $a (MiAaPQ)AAI29210084
035 $a AAI29210084
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Wang, Hao. $3 1911290
245 1 0 $a Information Theory for Trustworthy Machine Learning.
264 0 $c 2022
300 $a 1 online resource (246 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: Dissertations Abstracts International, Volume: 83-12, Section: B.
500 $a Advisor: Calmon, Flavio.
502 $a Thesis (Ph.D.)--Harvard University, 2022.
504 $a Includes bibliographical references
520 $a In this thesis, we provide an information-theoretic foundation for trustworthy machine learning (ML). ML algorithms are increasingly used in applications of significant social consequences. It is crucial to ensure that these algorithms are not just accurate but also generalizable, fair, and privacy-preserving. We develop new information-theoretic tools for proving rigorous performance guarantees for practical ML models and establishing protocols to ensure the responsible use of data.Towards rigorous performance guarantees, we derive generalization bounds for understanding the behaviors of complex ML models (e.g., neural networks). In particular, we study how the interaction between data distributions and optimization methods influences generalization. We consider a family of optimization methods---noisy iterative algorithms---and investigate their generalization capability. We derive distribution-dependent generalization bounds by connecting noisy iterative algorithms with additive noise channels found in information theory. Numerical experiments demonstrate that our results can help understand many empirical observations of neural networks. Towards responsible use, we characterize a fundamental limit of algorithmic fairness and privacy. Specifically, we provide conditions that ensure fair use of group attributes and analyze the "best" privacy-utility trade-offs among all privacy-preserving mechanisms. These theoretical results, in turn, inspire new algorithms that can repair unfair models or preserve privacy in data sharing. Finally, we evaluate the robustness of information-theoretic privacy measures and establish statistical consistency of "optimal" privacy mechanisms.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2023
538 $a Mode of access: World Wide Web
650 4 $a Applied mathematics. $3 2122814
650 4 $a Artificial intelligence. $3 516317
650 4 $a Information science. $3 554358
653 $a Information theory
653 $a Machine Learning
653 $a Data distribution
653 $a Privacy mechanisms
655 7 $a Electronic books. $2 lcsh $3 542853
690 $a 0364
690 $a 0800
690 $a 0723
710 2 $a ProQuest Information and Learning Co. $3 783688
710 2 $a Harvard University. $b Engineering and Applied Sciences - Applied Math. $3 3184332
773 0 $t Dissertations Abstracts International $g 83-12B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29210084 $z click for full text (PQDT)