CPU-GPU implementation of ensemble clustering algorithm for increased performance

Ensemble clustering algorithm is frequently used in machine learning algorithms, but it is also one of the most computationally intensive components thereby limiting their scalability. Parallel implementations of these algorithms enable researchers to pose bigger questions using larger datasets. The combinatorial and multi-objective nature of ensemble clustering algorithm makes running a large number of ensemble models a time-consuming task. In this paper, we present a CPU-GPU based implementation of ensemble clustering algorithm. Ensemble consists of agglomerative, gmm, kmeans, kmedians, and spectral clustering algorithms. Our implementation shows a 130x speedup over CPU only implementation. Primary usage and datasets used in our research are derived from risky decision making (RDM) domain.