The article explores the concept of decentralized federated learning, a rapidly evolving method that offers privacy-preserving features. However, this method also provides new attack surfaces for malicious users, potentially threatening model performance and user/data privacy. The study discusses possible variations of threats and adversaries in this learning method while also considering potential defense mechanisms. The trustability and verifiability of decentralized federated learning are also considered.
Publication date: 1 Feb 2024
Project Page: https://ieeexplore.ieee.org/document/9414644
Paper: https://arxiv.org/pdf/2401.17319
