论文标题
在最小冗余，分散优化的拜占庭断层耐受性
Byzantine Fault-Tolerance in Decentralized Optimization under Minimal Redundancy
论文作者
论文摘要
本文考虑了多代理分散化优化中拜占庭断层耐受性的问题。在此问题中，每个代理都有局部成本函数。分散优化算法的目的是允许代理合作计算其总成本函数的共同最小点。我们考虑了一定数量的代理可能是拜占庭式有缺陷的情况。这种错误的代理可能不会遵循规定的算法，并且它们可能与其他非故障代理人共享任意或不正确的信息。这种拜占庭药物的存在使典型的分散优化算法无效。我们提出了一种分散的优化算法，只要非故障代理具有最小的冗余，对有限数量的拜占庭式药物具有可证明的精确断层耐受性。
This paper considers the problem of Byzantine fault-tolerance in multi-agent decentralized optimization. In this problem, each agent has a local cost function. The goal of a decentralized optimization algorithm is to allow the agents to cooperatively compute a common minimum point of their aggregate cost function. We consider the case when a certain number of agents may be Byzantine faulty. Such faulty agents may not follow a prescribed algorithm, and they may share arbitrary or incorrect information with other non-faulty agents. Presence of such Byzantine agents renders a typical decentralized optimization algorithm ineffective. We propose a decentralized optimization algorithm with provable exact fault-tolerance against a bounded number of Byzantine agents, provided the non-faulty agents have a minimal redundancy.
