Two algorithms for distributed mode computing based on blended dynamics approach

This paper studies the distributed mode computing problem in a multi-agent system, in which each individual agent possesses a certain attribute and the agent group aims to agree upon the mode (the most frequent attribute owned by the agents) via distributed computing. Two algorithms are proposed, the first one estimates the frequency of all attributes at every agent, and then identifies the most frequent attribute as the mode; the second is based on a distributed consensus protocol that renders all the agents agreeing on an attribute whose frequency is no less than a given threshold. This protocol is then used as the main building block to compute the mode via a branch-and-bound algorithm. Analysis of both algorithms establishes finite time convergence and is based on the blended dynamics approach.