arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2002.03269math.OCcs.MA
收藏

分布式随机最优化的方差缩减线性收敛算法

Linearly Convergent Algorithm with Variance Reduction for Distributed Stochastic Optimization

Jinlong Lei, Peng Yi, Jie Chen, Yiguang Hong

摘要:

本文考虑了一个分布式随机强凸优化问题,其中连接在一个网络上的多个智能体的目标是合作地最小化所有智能体的局部费用函数的平均值。由于梯度估计的随机性和局部目标的分布性,目前还没有快速线性收敛的分布式算法。本文提出了一种新的分布式随机梯度跟踪算法,该算法通过增加采样梯度的批量大小来估计局部梯度。与一个无向连接的通信图和一个几何增加的批量大小,迭代收敛的平均几何速度(实现线性收敛)的最优解。建立了获得$ε$ -最优解的迭代、通信和预言复杂性。特别地,通信复杂度是$\mathcal{O}(\ln (1/ε))$,而预言复杂度(采样梯度的数量)是$\mathcal{O}(1/ε^2)$,其与集中式方法的数量级相同。因此,所提出的方案是通信高效的,而不需要额外的采样梯度。通过数值模拟验证了理论结果。

英文摘要:

This paper considers a distributed stochastic strongly convex optimization, where agents connected over a network aim to cooperatively minimize the average of all agents' local cost functions. Due to the stochasticity of gradient estimation and distributedness of local objective, fast linearly convergent distributed algorithms have not been achieved yet. This work proposes a novel distributed stochastic gradient tracking algorithm with variance reduction, where the local gradients are estimated by an increasing batch-size of sampled gradients. With an undirected connected communication graph and a geometrically increasing batch-size, the iterates are shown to converge in mean to the optimal solution at a geometric rate (achieving linear convergence). The iteration, communication, and oracle complexity for obtaining an $ε$-optimal solution are established as well. Particulary, the communication complexity is $\mathcal{O}(\ln (1/ε))$ while the oracle complexity (number of sampled gradients) is $\mathcal{O}(1/ε^2)$, which is of the same order as that of centralized approaches. Hence, the proposed scheme is communication-efficient without requiring extra sampled gradients. Numerical simulations are given to demonstrate the theoretic results.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑