Photo by hajjidirir from unsplash
Federated learning, as a novel paradigm of machine learning, is facing a series of challenges such as efficiency, privacy and robustness. The recently proposed EF-DP- SIGNSGD provides theoretical privacy protection… Click to show full abstract
Federated learning, as a novel paradigm of machine learning, is facing a series of challenges such as efficiency, privacy and robustness. The recently proposed EF-DP- SIGNSGD provides theoretical privacy protection for SIGNSGD with majority vote but weakens the capability to resist Byzantine attacks to some extent. To overcome this shortcoming and further greatly improve the communication efficiency, a new method called PCS-DP- SIGNSGD is proposed via using parallel compressed sensing. Simulation and analysis demonstrate that compared with EF-DP- SIGNSGD, PCS-DP- SIGNSGD can match or even improve the accuracy and enjoy stronger Byzantine robustness with 50% to 80% improvement in the uplink communication efficiency.
Journal Title: IEEE Communications Letters
Year Published: 2022
Link to full text (if available)
Share on Social Media: Sign Up to like & get
recommendations!