Methodology of Mean Shift Clustering Algorithm Implementation Based on Dataflow Computer

摘要

The article discusses the methodology development for implementing on a dataflow computer the Mean shift clustering algorithm, namely its subtype - Mean shift with flat core, also called FOREL (Formal Element). We have formalized the Mean shift algorithm for the dataflow implementation. We have also developed architecture of dataflow computer, identified the types of execution units, formulated algorithms of their operation and the information exchanging. The proposed computing methodology allows to reduce information traffic in the dataflow computing system for solving the clustering problem by combining a set of points located in the features space into a computational grid and reduce the time of clusters finding by parallelizing calculations. The methodology provides finding several clusters in the linear metric space, the number of which is unknown in advance due to the convergence of the Mean Shift algorithm.