报告人：罗珞 (复旦大学)
报告题目：On the Complexity of Decentralized Convex Optimization
摘要：We propose a decentralized convex optimization method with the (near) optimal computation rounds and communication rounds. We show the computation complexity can be further improved in the partial participation framework. For the finite-sum local functions, we allow different nodes establish their stochastic local gradient estimators with different mini-batch sizes in per iteration, which characterizes the heterogeneity and leads to the sharper local incremental first-order oracle complexity than the state-of-the-art methods.
