Massively Parallel Fourier-Space Cross-Correlation for Analyzing Highly Dimensional Time Series Databases

Lee, Matthias

Automated time series matching has recently become a very active field, producing many new methods of searching and matching time series. However, most of these algorithms and methods have not incorporated the immense power of parallel processing and the revolution of General Purpose computing on Graphics Processing Units (GPGPU). Most algorithms found in literature do not scale well over a massively parallel platforms, therefore we present a simple yet extremely effective and powerful method for searching massive time series databases by harnessing the parallel power of NVIDIA GPUs. We propose a parallel Fourier-space cross-correlation in connection with a super wide tree structure which we refer to as a Bonsai tree. Our massively parallel approach provides us with sub-second search times for multi-million-sized time series databases.

Return to oral presentation list