Analysis and design of a distributed k-winners-take-all model

The k-winners-take-all (kWTA) problem is to find the k largest inputs from N inputs. In this paper, we design and propose a novel distributed kWTA model, for which no central unit is needed to realize the computation of the k winners. As a result, the proposed model has the general advantages of distributed models over centralized ones, such as better robustness to faults of agents. The global asymptotic convergence of the proposed distributed model is proven. Besides, two numerical examples on networks of agents with static inputs and time-varying inputs are presented to validate the performance of the proposed model.