… Byzantine-robust federatedlearning (… poisoning attacks to Byzantine-robust federated learning. Existing studies [9, 66] only showed local model poisoningattacks to federatedlearning …
D Cao, S Chang, Z Lin, G Liu… - 2019 IEEE 25th …, 2019 - ieeexplore.ieee.org
… federatedlearning intuitively. In this paper, through real implementation of a federatedlearning system and distributed poisoningattacks… poisoned training samples, attackers, and attack …
L Lyu, H Yu, Q Yang - arXiv preprint arXiv:2003.02133, 2020 - arxiv.org
… Federatedlearning (FL) has recently emerged as a promising solution under this new reality… and two major attacks on FL: 1) poisoningattacks and 2) inference attacks, this paper …
… In this work, we systematized the threat models of poisoningattacks on federatedlearning (… and data poisoningattacks on FL (including existing and our improved attacks). Using these …
X Zhou, M Xu, Y Wu, N Zheng - Future Internet, 2021 - mdpi.com
… Table 1 summarizes the most notable poisoningattacks recent years. In this paper, we propose a model poisoningattack in federatedlearning that is based on optimization, which is …
C Fung, CJM Yoon, I Beschastnikh - arXiv preprint arXiv:1808.04866, 2018 - arxiv.org
… variety of attacks, including model poisoning, … federatedlearning to sybil-based poisoning attacks. We then describe FoolsGold, a novel defense to this problem that identifies poisoning …
J Zhang, B Chen, X Cheng, HTT Binh… - IEEE Internet of Things …, 2020 - ieeexplore.ieee.org
… Hence, we propose a novel generative poisoningattack model against federatedlearning by … can successfully launch the poisoningattack under a more practical threat assumption. …
… during the federated training process. In this paper, we present a general framework for model poisoningattacks on FL. We show that our framework leads to poisoningattacks that …
S Awan, B Luo, F Li - Computer Security–ESORICS 2021: 26th European …, 2021 - Springer
… To obtain a fair understanding of the defenses’ performance under a more realistic setting, we adopt a federatedlearning system with 100 clients and the MNIST dataset with 10 classes, …