國立虎尾科技大學 |

Vulnerability and Robustness in Artificial Intelligence for Complex Networks.

紀錄類型:	書目-語言資料,手稿 : Monograph/item
正題名/作者:	Vulnerability and Robustness in Artificial Intelligence for Complex Networks./
作者:	Miller, Benjamin A.
面頁冊數:	1 online resource (230 pages)
附註:	Source: Dissertations Abstracts International, Volume: 85-02, Section: B.
Contained By:	Dissertations Abstracts International85-02B.
標題:	Computer science. -
電子資源:	click for full text (PQDT)
ISBN:	9798380121187

Vulnerability and Robustness in Artificial Intelligence for Complex Networks.
Miller, Benjamin A.

Vulnerability and Robustness in Artificial Intelligence for Complex Networks. - 1 online resource (230 pages)

Source: Dissertations Abstracts International, Volume: 85-02, Section: B.

Thesis (Ph.D.)--Northeastern University, 2023.

Includes bibliographical references

Artificial intelligence (AI) is being adopted widely across society and has the potential to greatly improve human productivity. The methods, however, are far from perfect and there have been some high-profile failures. In an adversarial context, where a malicious actor actively attempts to mislead the user, the risk of such a situation increases. In high-stakes application areas, such as defense, making these methods robust to attack is critical to achieving their full potential.In this dissertation, I consider vulnerability and robustness to attack in three particular AI tasks with respect to graphs: vertex classification, community detection, and shortest path computation. My first study focuses on exploiting features of complex networks to make recent structural attacks against graph convolutional network-based vertex classifiers more costly. I demonstrate that these features are useful in several real networks and across various synthetic graphs, exploring the effects of homophily, degree heterogeneity, clustering, and attribute information value. I consider a similar problem in an unsupervised context with community detection, where my investigation shows that using overlapping community detection methods makes it much more difficult for a node to move out of a community of interest through adding edges alone. Turning my attention to network navigation, I consider an attacker that can remove edges to make a particular path be the shortest between two nodes. I prove that an attacker cannot efficiently find the optimal attack in general, but present the PATHATTACK algorithm, which guarantees an approximate solution and often does identify the optimal attack. In addition, PATHATTACK can be used in a heuristic method when the attacker targets a node or edge, rather than a specific path. Finally, I consider defending against shortest path attacks by modifying edge weights. The optimal defense also cannot be found efficiently, but I propose a heuristic method called PATHDEFENSE that uses the multiplicity of paths between nodes to increase the attacker's burden. Results across all three areas show that a defender can leverage network structure to make attacks more costly.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798380121187Subjects--Topical Terms:

573171
Computer science.
Subjects--Index Terms:

Network ScienceIndex Terms--Genre/Form:

554714
Electronic books.

Vulnerability and Robustness in Artificial Intelligence for Complex Networks.
LDR:03515ntm a22003617 4500 001 1148717
005 20240930100125.5
006 m o d
007 cr bn ---uuuuu
008 250605s2023 xx obm 000 0 eng d
020 $a 9798380121187
035 $a (MiAaPQ)AAI30635646
035 $a AAI30635646
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Miller, Benjamin A. $3 1474756
245 1 0 $a Vulnerability and Robustness in Artificial Intelligence for Complex Networks.
264 0 $c 2023
300 $a 1 online resource (230 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: 85-02, Section: B.
500 $a Advisor: Eliassi-Rad, Tina.
502 $a Thesis (Ph.D.)--Northeastern University, 2023.
504 $a Includes bibliographical references
520 $a Artificial intelligence (AI) is being adopted widely across society and has the potential to greatly improve human productivity. The methods, however, are far from perfect and there have been some high-profile failures. In an adversarial context, where a malicious actor actively attempts to mislead the user, the risk of such a situation increases. In high-stakes application areas, such as defense, making these methods robust to attack is critical to achieving their full potential.In this dissertation, I consider vulnerability and robustness to attack in three particular AI tasks with respect to graphs: vertex classification, community detection, and shortest path computation. My first study focuses on exploiting features of complex networks to make recent structural attacks against graph convolutional network-based vertex classifiers more costly. I demonstrate that these features are useful in several real networks and across various synthetic graphs, exploring the effects of homophily, degree heterogeneity, clustering, and attribute information value. I consider a similar problem in an unsupervised context with community detection, where my investigation shows that using overlapping community detection methods makes it much more difficult for a node to move out of a community of interest through adding edges alone. Turning my attention to network navigation, I consider an attacker that can remove edges to make a particular path be the shortest between two nodes. I prove that an attacker cannot efficiently find the optimal attack in general, but present the PATHATTACK algorithm, which guarantees an approximate solution and often does identify the optimal attack. In addition, PATHATTACK can be used in a heuristic method when the attacker targets a node or edge, rather than a specific path. Finally, I consider defending against shortest path attacks by modifying edge weights. The optimal defense also cannot be found efficiently, but I propose a heuristic method called PATHDEFENSE that uses the multiplicity of paths between nodes to increase the attacker's burden. Results across all three areas show that a defender can leverage network structure to make attacks more costly.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Computer science. $3 573171
653 $a Network Science
653 $a Community detection methods
653 $a PATHATTACK algorithm
655 7 $a Electronic books. $2 local $3 554714
690 $a 0984
690 $a 0800
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a Northeastern University. $b Interdisciplinary. $3 1474757
773 0 $t Dissertations Abstracts International $g 85-02B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30635646 $z click for full text (PQDT)