Density-Based Clustering for Knowledge Discovery of High-Dimensional Time Series Data using FPGAs

Density-Based Clustering is a form of Unsupervised Machine Learning where the concentration of data is used to define groups of similar instances. Density-Based Clustering has the advantage that it supports arbitrary cluster shapes and sizes for a relatively low computational cost. This characteristic allows the data to clearly define clusters and provides a good approach for Knowledge Discovery particularly with high-dimensional data. In this paper, we consider the use of Density-Based Clustering as an approach for the challenging task of Knowledge Discovery of high-dimensional time series data. Specifically, in this paper we leverage this type of clustering to provide insight into time series data. Time series data, and particularly high-dimensional time series data, can be very challenging to gain insight and discover structure in data. With the added dimension of time we also need to consider the stationarity of clusters with respect to density and location of clusters as well as the time order vectors of clusters, instances and relative change across dimensions. Furthermore, this paper considers implementation of Density-Based 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 Density-Based Clustering of high-dimensional time series data. Such an approach supports fast Machine Learning in an environment of low stationarity. Finally, an example is provided based on the Xilinx Adaptive Compute Acceleration Platform (ACAP) using a Xilinx Versal Premium Series device to illustrate the concepts in this paper.