Security and Privacy Threats to Federated Learning: Issues, Methods, and Challenges

Abstract

Federated learning (FL) has nourished a promising method for data silos, which enables multiple participants to construct a joint model collaboratively without centralizing data. The security and privacy considerations of FL are focused on ensuring the robustness of the global model and the privacy of participants’ information. However, the FL paradigm is under various security threats from the adversary aggregator and participants. Therefore, it is necessary to comprehensively identify and classify potential threats to provide a theoretical basis for FL with security guarantees. In this paper, a unique classification of attacks, which reviews state-of-the-art research on security and privacy issues for FL, is constructed from the perspective of malicious threats based on different computing parties. Specifically, we categorize attacks with respect to performed by aggregator and participant, highlighting the Deep Gradients Leakage attacks and Generative Adversarial Networks attacks. Following an overview of attack methods, we discuss the primary mitigation techniques against security risks and privacy breaches, especially the application of blockchain and Trusted Execution Environments. Finally, several promising directions for future research are discussed.

Federated Learning Based Privacy Ensured Sensor Communication in IoT Networks: A Taxonomy, Threats and Attacks

“…The classification of attacks in these systems based on a more comprehensive viewpoint is shown in Figure 13. The authors in [22] have classified attacks in three broad categories: (I) Attacks focused on data [26], [89], (ii) Attacks focused on algorithm [89], and (iii) Attacks focused on federation. These are briefly explained as follows:…”

“…This is due to the fact that adversarial FL clients have the ability to manipulate and shift the boundaries of the model while it is being developed [90]. The attacks that are focused on data is categorized into three types namely: poisoning attack [22], [26], [27], backdoor attacks, and evasion attacks [22]. In poisoning attack, the adversaries alter the behavior of the target FL model.…”

“…Secret Sharing refers to the distribution of a secret among a group of people in such a way that no single person has access to any information about the secret that can be understood; however, if enough people combine their "shares," the secret can be reconstructed [23], [26]. SMC is a cryptographic method that enables distant users to jointly compute an objective function without disclosing their data.…”

Analysis of the Factors Influencing the Predictive Learning Performance Using Federated Learning

“…Furthermore, we offer an extensive examination of the present challenges faced by FL, encompassing aspects such as safeguarding user privacy, communication expenses, system heterogeneity, and the lack of reliability in model uploads Analysis of the Factors Influencing the Predictive Learning Performance Using Federa [17] Currently, Florida is encountering a range of challenges, a few of which will be discussed herein. Subsequently, examining the numerous potential avenues for future research will ensue.…”

A Survey on Challenges of Federated Learning

“…When many devices can access the server, one must consider that it opens doors for different attacks. These attacks may be targeted on the server or the global model (Zhang, J., Zhu, H., Wang, F., Zhao, J., Xu, Q., & Li, H. 2022). Some of these attacks are very hard to detect and, when not prevented, can drastically decrease the global model's performance.…”