Design space exploration of the KNN imputation on FPGA
Ahmad Al-Zoubi, Konstantinos Tatas, Costas Kyriacou · 2018
The K-Nearest-Neighbor (KNN) imputation algorithm is regarded one of the most useful algorithms in data mining applications. Despite its wide acceptance in the research and industrial communities and its competitive efficiency, the KNN imputation high computation cost remains the main performance bottleneck, especially for large high-dimensional data sets. This work perform design space exploration on Field-Programmable-Gate-Array (FPGA) using Vivado HLS and propose a fixed-point arithmetic KNN imputation algorithm. The design has been implemented and tested with Yeast UCI dataset, and the experimental results demonstrate a clear speedup over the serial performance.