New Convergence Analysis of the BEER Algorithm in Decentralized Nonconvex Optimization

This paper presents an improved convergence proof for the BEER algorithm (Zhao et al., NeurIPS 2022) within the domain of decentralized nonconvex optimization. BEER undergoes a thorough reexamination accompanied by a novel proof, showcasing an enhanced convergence rate. Our research enriches the comprehension of BEER’s convergence characteristics, offering a compelling advancement for researchers and practitioners actively involved in evolving communication-efficient or privacy-enhanced decentralized optimization algorithms, particularly in the context of Internet of Things (IoT), Beyond 5G (B5G), and 6G technologies.