A Markov-Bayesian Defense Against Backdoor Attacks in IoT Federated Learning
Open Access DepositedMB-Weight
MB-Weight
especially backdoor attacks are pernicious (Wang et al., 2024). Several defense mechanisms have been proposed against attacks in FL, but the problem is they are not adaptable. Because most protections focus on known attack patterns, they are weak against training-time poisoning that activates malicious behavior under specific conditions. This shortcoming threatens both security and reliability. To address this, we propose MB-Weight, a novel FL defense mechanism that models client uncertainty with Bayesian inference, tracks temporal reputation with a Markov chain to spot and curb malicious updates, and uses trust-weighted aggregation so risky clients have minimal influence on the global model.
A Markov-Bayesian Defense Against Backdoor Attacks in IoT Federated Learning The Internet of Things (IoT) represents a significant paradigm shift within the technological domain, enabling extensive interconnection of devices and sensors, thereby advancing the exchange of information through advanced network architectures. This revolution has created many new uses but also introduces unique challenges in security and data privacy. To address the critical concern of privacy in IoT devices, we are investigating the potential of Federated Learning (FL) as a viable solution. It is a method to train machine learning models without sharing all the data. Like McMahan et al. (2017) highlighted, each device makes updates to the main model and sends only those updates to the server. This approach safeguards data privacy and optimizes communication bandwidth utilization. But federated learning still has security issues
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.