数学科学学院学术报告
Improving the Worst-Case Bidirectional Communication Complexity for Nonconvex Distributed Optimization
Kaja Gruntkowska
（阿卜杜拉国王科技大学）
报告时间：2025年4月21日 星期一 下午 16：00-17:00
报告地点：沙河校区E806
报告摘要：Effective communication between the server and workers plays a key role in distributed optimization. In this paper, we focus on optimizing the server-to-worker communication, uncovering inefficiencies in prevalent downlink compression approaches. Considering first the pure setup where the uplink communication costs are negligible, we introduce MARINA-P, a novel method for downlink compression, employing a collection of correlated compressors. Theoretical analysis demonstrates that MARINA-P with permutation compressors can achieve a server-to-worker communication complexity improving with the number of workers, thus being provably superior to existing algorithms. We further show that MARINA-P can serve as a starting point for extensions such as methods supporting bidirectional compression. We introduce M3, a method combining MARINA-P with uplink compression and a momentum step, achieving bidirectional compression with provable improvements in total communication complexity as the number of workers increases. Theoretical findings align closely with empirical experiments, underscoring the efficiency of the proposed algorithms.
报告人简介：Kaja Gruntkowska is a PhD student in Optimization for Machine Learning at KAUST, advised by Prof. Peter Richtárik. Her research focuses on developing the algorithmic and mathematical foundations of randomized optimization, with a particular emphasis on distributed computing. She works on designing practically motivated algorithms with provable convergence guarantees, bridging theory and real-world applications to advance scalable machine learning. She completed her Bachelor's in Mathematics and Statistics at the University of Warwick and earned a Master's in Statistical Science from the University of Oxford.
邀请人：谢家新
