Binomial Distribution based K-means for Graph Partitioning Approach in Partially Reconfigurable Computing system

• Published in: Iranian Conference on Electrical Engineering pp. 568 - 572
• Main Authors: Asgari, Zahra, Mastoori, Maryam Sadat
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.05.2021
• Subjects: Bin packing; Binomial Distribution based K-means; Classification algorithms; Clustering algorithms; Elbow; Electrical engineering; Graph partitioning algorithms; K-means algorithm; Partitioning algorithms; Q-factor; Reconfigurable computing; Simulated annealing; Unsupervised clustering
• ISSN: 2642-9527
• DOI: 10.1109/ICEE52715.2021.9544358

Graph partitioning algorithms have been utilized to execute complex applications, where there is no enough space to run the whole application once, like in limited reconfigurable computing resources. If we have found an "optimal" clustering of a data set, it can be proved that optimal partitioning can be achieved. K-means based algorithms are widely used to partition subjects where there is no information about the number of clusters. A vital issue in the mentioned method is how to define a good centroid, which has the principal role in "good" clustering. In this paper, we introduced a new way to determine purposive centroids, based on Binomial Distribution to reduce the risk of randomly seeds selection, Elbow Diagram to achieve the optimum number of clusters, and finally, Bin Packing to classify nodes in defined clusters with considering Utilization Factor (UF) due to the limited area of Run Space. The proposed algorithm, called Binomial Distribution based K-means (BDK), is compared with common graph partitioning algorithms like Simulated Annealing Algorithm (SA), Density K-means (DK), and a link elimination partitioning with different scenarios such as simple and complex applications. The concluding results show that the proposed algorithm decreases the error of partitioning by 24% compared to the other clustering techniques. On the other hand, the Quality Factor (QF) is increased 41% in this way. Execution Time (EX.T) to achieve the required number of clusters is reduced significantly.