Accelerating Dynamic Time Warping Subsequence Search with GPUs and FPGAs

Abstract

Abstract-Many time series data mining problems require subsequence similarity search as a subroutine. While this can be performed with any distance measure, and dozens of distance measures have been proposed in the last decade, there is increasing evidence that Dynamic Time Warping (DTW) is the best measure across a wide range of domains. Given DTW's usefulness and ubiquity, there has been a large community-wide effort to mitigate its relative lethargy. Proposed speedup techniques include early abandoning strategies, lower-bound based pruning, indexing and embedding. In this work we argue that we are now close to exhausting all possible speedup from software, and that we must turn to hardware-based solutions if we are to tackle the many problems that are currently untenable even with stateof-the-art algorithms running on high-end desktops. With this motivation, we investigate both GPU (Graphics Processing Unit) and FPGA (Field Programmable Gate Array) based acceleration of subsequence similarity search under the DTW measure. As we shall show, our novel algorithms allow GPUs, which are typically bundled with standard desktops, to achieve two orders of magnitude speedup. For problem domains which require even greater scale up, we show that FPGAs costing just a few thousand dollars can be used to produce four orders of magnitude speedup. We conduct detailed case studies on the classification of astronomical observations and similarity search in commercial agriculture, and demonstrate that our ideas allow us to tackle problems that would be simply untenable otherwise.

Fast spoken query detection using lower-bound Dynamic Time Warping on Graphical Processing Units

“…Figure 3 shows a comparison of the computation time for the two different aforementioned parallel DTW implementations. In the figure, the x-axis shows the number of DTWs computed, the red curve represents the time consumed for the method proposed by Sart et al [5], while the blue curve shows the method used in the current implementation. The figure indicates that both methods consumed similar time when computing less than 5,000 DTWs.…”

“…In the future, we plan to investigate additional GPU efficiencies, implementations using GPU farms, as well as other parallel computing architectures such as FPGAs [5].…”

“…In [5], the DTW GPU implementation was to use each thread in each kernel instance to compute the absolute distance matrix A as well as the DTW alignment. In order to compare the two approaches, we implemented both methods and report the results of the comparison in the next section.…”

“…[2,3,4] have demonstrated how GPUs can be used to rapidly calculate acoustic likelihoods for large mixture models. Sart et al provided a Dynamic Time Warping (DTW) search algorithm for GPUs in [5], claiming that they could speed up the run time by up to two orders of magnitude.…”

An advanced energy management framework to promote energy awareness

“…Figure 4 shows a modification of their algorithm with real-time Z-normalisation of the input data stream, allowing manufacturing energy data to be processed and converted into the set of events related to the operational state of a machine. This modification is necessary since the original algorithm does not normalize the offset or amplitude of the matching data sequences (Sart et al, 2010). Therefore, it is assumed that the data sequences will naturally have the same offset and amplitude as the matching reference pattern.…”