• 

  濮实

  香港中文大学（深圳）数据科学学院助理教授、优德官网 副研究员

  执行主席

  嘉宾先容

  演讲摘要
• 

  黄琨

  香港中文大学（深圳）数据科学学院博士生

  Distributed Random Reshuffling over Networks

  嘉宾先容

  黄琨于2018年获同济大学数学科学学院数学与应用数学学士学位，，，，，2020年获康涅狄格大学统计学硕士学位，，，，，现在在香港中文大学（深圳）数据科学学院攻读数据科学博士学位，，，，，曾在优德官网机械学习与应用中心担当研究助理。。。。。。。他的研究兴趣包括漫衍式优化。。。。。。。其事情曾在IEEE Transactions on Automatic Control揭晓。。。。。。。

  演讲摘要

  In this paper, we consider distributed optimization problems where n agents, each possessing a local cost function, collaboratively minimize the average of the local cost functions over a connected network. To solve the problem, we propose a distributed random reshuffling (D-RR) algorithm that invokes the random reshuffling (RR) update in each agent. We show that D-RR inherits favorable characteristics of RR for both smooth strongly convex and smooth nonconvex objective functions. In particular, for smooth strongly convex objective functions, D- RR achieves O(1/T²) rate of convergence (where T counts the epoch number) in terms of the squared distance between the iterate and the global minimizer. When the objective function is assumed to be smooth nonconvex, we show that D-RR drives the squared norm of the gradient to 0 at a rate of O(1/T^2/3). These convergence results match those of centralized RR (up to constant factors) and outperform the distributed stochastic gradient descent (DSGD) algorithm if we run a relatively large number of epochs. Finally, we conduct a set of numerical experiments to illustrate the efficiency of the proposed D-RR method on both strongly convex and nonconvex distributed optimization problems. 

• 

  邱俊文

  香港中文大学（深圳）数据科学学院博士生

  Convergence of Random Reshuffling under the KL Inequality

  嘉宾先容

  邱俊文于2019年获暨南大学信息治理与信息系统学士学位，，，，，现在在香港中文大学（深圳）数据科学学院攻读数据科学博士学位，，，，，曾在优德官网机械学习与应用中心担当研究助理。。。。。。。他的研究兴趣包括非凸随机优化。。。。。。。

  演讲摘要

  In this talk, we present novel convergence properties of the random reshuffling (RR) method for smooth nonconvex optimization problems with a finite-sum structure. Though this method is widely utilized in practice, e.g., in the training of neural networks, its convergence behavior is only understood in several limiting cases. Based on the Kurdyka-Lojasiewicz (KL) inequality, we verify that the whole sequence of iterates generated by RR is convergent and converges to a single stationary point of problem in an almost sure sense. Corresponding rates of convergence are derived which depend on the KL exponent and suitably selected diminishing step sizes. When the KL exponent lies in $[0,\frac12]$, the convergence is at a rate of $\mathcal{O}(t^{-1})$ with $t$ counting the number of iterations. The standard KL inequality-based convergence framework only applies to algorithms with a certain descent property. We conduct a novel analysis for the non-descent RR method with diminishing step sizes based on the KL inequality which generalizes the standard KL framework and techniques.

视频回首