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

Crafting Machine Learning Defenses Against Adversaries.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Crafting Machine Learning Defenses Against Adversaries./
Author:	Park, Won.
Description:	1 online resource (137 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 84-12, Section: B.
Contained By:	Dissertations Abstracts International84-12B.
Subject:	Computer science. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30548379click for full text (PQDT)
ISBN:	9798379565121

Crafting Machine Learning Defenses Against Adversaries.
Park, Won.

Crafting Machine Learning Defenses Against Adversaries. - 1 online resource (137 pages)

Source: Dissertations Abstracts International, Volume: 84-12, Section: B.

Thesis (Ph.D.)--University of Michigan, 2023.

Includes bibliographical references

Machine learning systems are becoming widely adopted and ubiquitous. Not only are there a growth of products in which machine learning is at their core like autonomous vehicles, but even traditional companies in fields such as finance, telecommunications, and travel are integrating machine learning into their internal structure.However, like any system, machine learning platforms are prone to security risks and vulnerabilities. Coupled with an ever-accelerating deployment and usage of machine learning systems, the attackers' chance of success and capability of damage increases just as rapidly. What is especially concerning is the large surface area of the machine learning pipeline that is available for attack - from training all the way to inference. With such a wide variety of attack combinations possible, there remains a need to address and explore the many types of attacks and defense that are possible in a machine learning environment.To address this goal, in this dissertation, we explore some of the different types of security vulnerabilities and attacks that are possible with different types of machine learning systems. At the inference level, my dissertation explores the possibility of crafting adversarial examples on multimodal sensor fusion models - the kind that would be used by autonomous vehicle manufacturers. We also explore a new technique that can be used for defending against adversarial examples: adversarial fine-tuning. Our proposed methodology exceeds the state of the art in terms of robust accuracy and benign accuracy, while still taking much faster to train than traditional adversarial training.We also explore a gap in the study of attacks during the training phase of the model (i.e. poisoning or backdoor attacks), by exploring the frequency domain of images and how that could affect attacks and detection defenses.Finally, through a collaboration at Ericsson, Inc., we explore how a machine learning framework can be deployed to detect anomalous data while still being cognizant of industry restrictions and metrics.

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

Mode of access: World Wide Web

ISBN: 9798379565121Subjects--Topical Terms:

523869
Computer science.
Subjects--Index Terms:

Machine learningIndex Terms--Genre/Form:

542853
Electronic books.

Crafting Machine Learning Defenses Against Adversaries.
LDR:03466nmm a2200409K 4500 001 2362465
005 20231027104047.5
006 m o d
007 cr mn ---uuuuu
008 241011s2023 xx obm 000 0 eng d
020 $a 9798379565121
035 $a (MiAaPQ)AAI30548379
035 $a (MiAaPQ)umichrackham004924
035 $a AAI30548379
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Park, Won. $3 3174268
245 1 0 $a Crafting Machine Learning Defenses Against Adversaries.
264 0 $c 2023
300 $a 1 online resource (137 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: 84-12, Section: B.
500 $a Advisor: Mao, Z. Morley.
502 $a Thesis (Ph.D.)--University of Michigan, 2023.
504 $a Includes bibliographical references
520 $a Machine learning systems are becoming widely adopted and ubiquitous. Not only are there a growth of products in which machine learning is at their core like autonomous vehicles, but even traditional companies in fields such as finance, telecommunications, and travel are integrating machine learning into their internal structure.However, like any system, machine learning platforms are prone to security risks and vulnerabilities. Coupled with an ever-accelerating deployment and usage of machine learning systems, the attackers' chance of success and capability of damage increases just as rapidly. What is especially concerning is the large surface area of the machine learning pipeline that is available for attack - from training all the way to inference. With such a wide variety of attack combinations possible, there remains a need to address and explore the many types of attacks and defense that are possible in a machine learning environment.To address this goal, in this dissertation, we explore some of the different types of security vulnerabilities and attacks that are possible with different types of machine learning systems. At the inference level, my dissertation explores the possibility of crafting adversarial examples on multimodal sensor fusion models - the kind that would be used by autonomous vehicle manufacturers. We also explore a new technique that can be used for defending against adversarial examples: adversarial fine-tuning. Our proposed methodology exceeds the state of the art in terms of robust accuracy and benign accuracy, while still taking much faster to train than traditional adversarial training.We also explore a gap in the study of attacks during the training phase of the model (i.e. poisoning or backdoor attacks), by exploring the frequency domain of images and how that could affect attacks and detection defenses.Finally, through a collaboration at Ericsson, Inc., we explore how a machine learning framework can be deployed to detect anomalous data while still being cognizant of industry restrictions and metrics.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2023
538 $a Mode of access: World Wide Web
650 4 $a Computer science. $3 523869
650 4 $a Information technology. $3 532993
653 $a Machine learning
653 $a Adversaries
653 $a Network security
653 $a Network attacks
653 $a Adversarial fine-tuning
655 7 $a Electronic books. $2 lcsh $3 542853
690 $a 0984
690 $a 0489
690 $a 0800
710 2 $a ProQuest Information and Learning Co. $3 783688
710 2 $a University of Michigan. $b Computer Science & Engineering. $3 3285590
773 0 $t Dissertations Abstracts International $g 84-12B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30548379 $z click for full text (PQDT)