Multiprocessor Architecture For Fuzzy-C-Mean Clustering Algorithm For Edge Computing

Abstract

Cloud-based systems have become very popular these days, which consists of a server and node devices. These node devices are called edge devices or edge of the system. With ever increasing size of data and complex algorithms, the load on servers have increased many folds. Hence there is a requirement of shifting this load to the edge devices. Shifting load from server to edge nodes is called edge computing. Now Fuzzy C-Means (FCM) is an unsupervised machine learning algorithm that is used for data clustering. It is also called soft k-means or soft clustering as each data point can belong to more than one cluster. In cloud-based applications it can affect the performance of server due to its complexity. In edge computing systems workload is put close to the edge where data is being created, this helps improve response time and save bandwidth. FCM can be incorporated in various smart nodes-based devices by assigning data to relative clusters. Image segmentation is one of such applications that can be implemented using FCM. In proposed design FCM is implemented using multicore architecture, which consists of P processing units called Tiles. Each Tile process a chunk of data in parallel and final results are calculated by sharing of data between the Tiles. This improves the overall performance of the system by speed up as compared to the traditional sequential architecture. The proposed architecture can be used to design a smart edge computing-based system with high performance and better throughput.