针对多维凸分布式优化中的拜占庭攻击问题,提出含距离过滤与逐分量过滤的容错算法,可在未知攻击者身份时容忍每节点至多 F 个恶意邻居并保证渐近收敛至全局最优附近。
摘要:
分布式优化问题要求一组智能体利用从邻居处接收到的信息,就一个使其局部代价函数平均值最小化的参数达成一致。尽管存在多种能够解决该问题的分布式优化算法,但它们通常容易受到不遵守算法的恶意(即“拜占庭”)智能体的攻击。近期为解决这一问题所做的尝试主要集中于一维函数,或者在关于各智能体处函数统计特性的某些假设下提供分析。本文提出了一种适用于多维凸函数的容错分布式优化算法。我们的方案在算法的每次迭代中包含两个过滤步骤:(1)基于距离的极端状态移除;(2)逐分量的极端状态移除。我们证明,该算法能够在每个常规节点的邻居中缓解至多 F 个拜占庭智能体的影响,而无需事先知道拜占庭智能体的身份。特别地,我们表明,如果网络拓扑满足某些条件,则所有常规状态都保证渐近收敛到一个包含全局最小化点的有界区域。
英文摘要:
The problem of distributed optimization requires a group of agents to reach agreement on a parameter that minimizes the average of their local cost functions using information received from their neighbors. While there are a variety of distributed optimization algorithms that can solve this problem, they are typically vulnerable to malicious (or ``Byzantine'') agents that do not follow the algorithm. Recent attempts to address this issue focus on single dimensional functions, or provide analysis under certain assumptions on the statistical properties of the functions at the agents. In this paper, we propose a resilient distributed optimization algorithm for multi-dimensional convex functions. Our scheme involves two filtering steps at each iteration of the algorithm: (1) distance-based and (2) component-wise removal of extreme states. We show that this algorithm can mitigate the impact of up to $F$ Byzantine agents in the neighborhood of each regular node, without knowing the identities of the Byzantine agents in advance. In particular, we show that if the network topology satisfies certain conditions, all of the regular states are guaranteed to asymptotically converge to a bounded region that contains the global minimizer.
Resilient Decentralized Preconditioned Stochastic Proximal Optimization with Gradient Sketching over Time-Varying Networks
Hubei Normal University(湖北师范大学);City University of Hong Kong(香港城市大学);University of New South Wales(新南威尔士大学);Southwest University(西南大学);Shenzhen University of Advanced Technology(深圳先进技术研究院)
Prox-DBRO-VR: A Unified Analysis on Byzantine-Resilient Decentralized Stochastic Composite Optimization with Variance Reduction and Non-Asymptotic Convergence Rates
Central South University(中南大学);City University of Hong Kong(香港城市大学);University of New South Wales(新南威尔士大学);Southwest University(西南大学)
Byzantine-Robust Distributed SGD: A Unified Analysis and Tight Error Bounds
School of Advanced Interdisciplinary Sciences, University of Chinese Academy of Sciences(中国科学院大学前沿交叉科学学院);School of Mathematical Sciences, University of Chinese Academy of Sciences(中国科学院大学数学科学学院);Ministry of Education Key Laboratory of Mathematics and Information Networks, School of Mathematical Sciences, Beijing University of Posts and Telecommunications(北京邮电大学数学科学学院数学与信息网络教育部重点实验室)
在大规模分布式学习框架(如联邦学习)中,非凸优化的鞍点避免问题具有相当大的挑战性,特别是在拜占庭工人存在的情况下。\cite{nest}中著名的正则化牛顿法是在标准集中式(非分布式)设置中避免鞍点的最优雅的方法之一。在本文中,我们扩展的正则化牛顿方法的分布式框架,同时解决一些实际的挑战,如通信瓶颈和拜占庭攻击。注意,在拜占庭机器存在的情况下,鞍点避免的问题变得更加关键,因为流氓机器可能在损失函数的鞍点附近创建\emph{fake local minima},也称为鞍点攻击。作为一个二阶算法,我们的迭代复杂度比一阶算法低得多。此外,我们使用压缩(或稀疏化)技术,如$δ$ -近似压缩通信效率。我们获得了理论保证,我们提出的计划下的几个设置,包括近似(二次采样)梯度和海森。此外,我们使用标准数据集和几种类型的拜占庭攻击的实验验证了我们的理论研究结果,并获得了改进的$25\%$相对于一阶方法的迭代复杂度。