The New Jersey Institute of Technology's
Electronic Theses & Dissertations Project

Title:	Un-fair trojan: Targeted backdoor attacks against model fairness
Author:	Furth, Nicholas
View Online:	njit-etd2022-001 (ix, 42 pages ~ 2.1 MB pdf)
Department:	Department of Electrical and Computer Engineering
Degree:	Master of Science
Program:	Computer Engineering
Document Type:	Thesis
Advisory Committee:	Khreishah, Abdallah (Committee chair) Wang, Cong (Committee member) Phan, Hai Nhat (Committee member) Liu, Qing Gary (Committee member)
Date:	2022-05
Keywords:	Machine learning Federated learning Fair machine learning Trojan backdoors Decentralized learning
Availability:	Unrestricted
Abstract:	Machine learning models have been shown to be vulnerable against various backdoor and data poisoning attacks that adversely affect model behavior. Additionally, these attacks have been shown to make unfair predictions with respect to certain protected features. In federated learning, multiple local models contribute to a single global model communicating only using local gradients, the issue of attacks become more prevalent and complex. Previously published works revolve around solving these issues both individually and jointly. However, there has been little study on the effects of attacks against model fairness. Demonstrated in this work, a flexible attack, which we call Un-Fair Trojan, that targets model fairness while remaining stealthy can have devastating effects against machine learning models.