Acceleration of the SPADE Method Using a Custom-Tailored FP-Growth Implementation.

Area of Science:

• Computational Neuroscience
• Bioinformatics
• High-Performance Computing

Background:

• The SPADE (spatio-temporal Spike PAtern Detection and Evaluation) method identifies recurring spatio-temporal patterns in neuronal spike trains.
• Long runtimes of the original SPADE method limit its application, especially with large datasets.

Purpose of the Study:

• To accelerate the SPADE method by optimizing its computationally intensive components: pattern mining and result filtering.
• To enable SPADE's application on diverse hardware, including low-power embedded systems.

Main Methods:

• Developed a customized, parallel, and distributed FP-Growth algorithm implementation tailored for SPADE.
• Evaluated the optimized implementation on a range of hardware, from traditional workstations to heterogeneous microservers and NVIDIA Jetson devices.

Main Results:

• The customized FP-Growth implementation accounts for 85-90% of the original SPADE runtime improvements.
• Achieved speedups of 27 to 200 times compared to the original SPADE implementation across different platforms.
• Reduced energy consumption by up to two orders of magnitude.

Conclusions:

• The optimized SPADE implementation offers substantial performance gains and energy efficiency.
• This advancement makes advanced spatio-temporal pattern analysis in neuronal activity more accessible and feasible on a wider array of computational platforms.