Ensemble Clustering for Unsupervised Learning of Time Series Data using FPGAs

Ensemble Clustering is a form of Unsupervised Machine Learning where several clustering algorithms are combined in an ensemble to improve performance over a single clustering algorithm. Outputs from several clustering algorithms are combined together to determine the final output of the ensemble. A consensus function is used for combining outputs from the clustering algorithms to improve output from the ensemble. Selecting a consensus function without any apriori knowledge is difficult, and many types of consensus functions exist with various levels of computational complexity. In this paper, we consider the use of Ensemble Clustering as an unsupervised learning approach to time series data. Specifically, in this paper the challenge is to extract key relationships in time series data, gain insight and discover structure in data. The paper considers the advantages and disadvantages of ensembles in unsupervised machine learning followed by a discussion of clustering algorithms. A discussion of consensus functions for ensembles of clustering algorithms is provided and followed by a discussion of unsupervised learning of time series data. Furthermore, this paper considers implementation of Ensemble Clustering using Field Programmable Gate Arrays (FPGAs). The approach described in this paper applies to high performance, high throughput and scalable implementations for Big Data. Design data is provided for FPGA implementation of Ensemble Clustering. Finally, an example is provided based on the Xilinx UltraScale+ FPGAs to illustrate the concepts in this paper.