A Case of Distributed Optimization in Adversarial Environment
A Case of Distributed Optimization in Adversarial Environment
复制标题
DOI：
10.1109/icassp.2019.8683442
复制
发表时间：
2019-05
期刊：
影响因子：
--
通讯作者：
Nikhil Ravi;A. Scaglione;A. Nedić
中科院分区：
文献类型：
--
作者：
Nikhil Ravi;A. Scaglione;A. Nedić
在本文中，我们考虑在包含常规节点和恶意节点（代理）的网络中解决分布式（基于共识）优化问题。常规节点正在执行分布式迭代算法来解决相关的优化问题，而恶意节点则注入虚假数据，目的是将迭代引导到符合其自身利益的点。该问题包括检测和隔离恶意代理，从而允许常规节点解决其优化问题。我们提出了一种使分布式优化算法上的数据注入攻击相形见绌的方法，该方法基于这样的想法：恶意节点（单独或协作）在广播消息时往往会暴露自己，意图使共识值远离网络中常规节点的最佳点。特别是，我们提供了一种新的基于梯度的度量来检测可能是恶意的邻居。我们还提供了一些模拟结果，证明了所提出方法的性能。
In this paper, we consider the problem of solving a distributed (consensus-based) optimization problem in a network that contains regular and malicious nodes (agents). The regular nodes are performing a distributed iterative algorithm to solve their associated optimization problem, while the malicious nodes inject false data with a goal to steer the iterates to a point that serves their own interest. The problem consists of detecting and isolating the malicious agents, thus allowing the regular nodes to solve their optimization problem. We propose a method to dwarf data injection attacks on distributed optimization algorithms, which is based on the idea that the malicious nodes (individually or in collaboration) tend to give themselves away when broadcasting messages with the intention to drive the consensus value away from the optimal point for the regular nodes in the network. In particular, we provide a new gradient-based metric to detect the neighbors that are likely to be malicious. We also provide some simulation results demonstrating the performance of the proposed approach.