Reputation-based Resilient Distributed Algorithms

Abstract

In this talk, the problem of designing distributed algorithms is presented and discussed its main challenges. Setting the task of having a distributed algorithm in a multi-agent system to achieve consensus of a variable of interest, we present a discrete-time reputation-based consensus algorithm that eliminates attacked agents' information for the case of a static topology and synchronous communication. Using only local information, agents can compute a reputation for each of its neighbors and assess the relative quality of the received data prior to the update. We show that, for each time step, the proposed method has quadratic complexity on the number of neighbors of each agent. Finally, we illustrate our method with several examples, portraying some attacking schemes successfully dealt by our method that succeed against the state-of-the-art.