Fast Quadratic Discriminant Analysis Using GPGPU for Sea Ice Forecasting

General Purpose computing on Graphics Processor Units (GPGPU) brings massively parallel computing (hundreds of compute cores) to the desktop at a reasonable cost, but requires that algorithms be carefully designed to take advantage of this power. The present work explores the possibilities of CUDA (NVIDIA Compute Unified Device Architecture) using GPGPU for Quadratic Discriminant (QD) analysis. QD analysis is a form of multivariate statistical analysis that can be applied to forecasting seasonal sea ice freeze-up and break-up. The forecast problem is formulated as a classification problem, with two classes (e.g., "ice" and "no ice") and the objective of the analysis is to decide which of the classes best describes the ice/no ice condition at a particular geographic point on a specified date. We have conducted experiments to measure the performance of the GPU with respect to the serial CPU, parallel CPU (OpenMP), MATLAB, MATLAB (Parallel for) implementations. The experiments consist of implementing a serial CPU, parallel CPU (OpenMP), MATLAB, MATLAB (Parallel for) and GPU versions of the QD analysis algorithm and executing all versions on several data sets to compare the performance. Our results show speed up of up to 426 times, reducing the elapsed time from over 15 hours to about 2 minutes.

[1]  Leon Reznik,et al.  GPU-based simulation of spiking neural networks with real-time performance & high accuracy , 2010, The 2010 International Joint Conference on Neural Networks (IJCNN).

[2]  Anil K. Jain,et al.  Feature Selection: Evaluation, Application, and Small Sample Performance , 1997, IEEE Trans. Pattern Anal. Mach. Intell..

[3]  Kevin Skadron,et al.  Scalable parallel programming , 2008, 2008 IEEE Hot Chips 20 Symposium (HCS).

[4]  Pat Hanrahan,et al.  Brook for GPUs: stream computing on graphics hardware , 2004, ACM Trans. Graph..

[5]  Jens H. Krüger,et al.  A Survey of General‐Purpose Computation on Graphics Hardware , 2007, Eurographics.

[6]  Michael Wolfe,et al.  Implementing the PGI Accelerator model , 2010, GPGPU-3.

[7]  David G. Stork,et al.  Pattern Classification , 1973 .

[8]  Michel Barlaud,et al.  Fast k nearest neighbor search using GPU , 2008, 2008 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops.

[9]  Anil K. Jain,et al.  39 Dimensionality and sample size considerations in pattern recognition practice , 1982, Classification, Pattern Recognition and Reduction of Dimensionality.

[10]  Carl Howell,et al.  Iceberg and ship detection and classification in single, dual and quad polarized synthetic aperture radar , 2008 .

[11]  Anil K. Jain,et al.  Small Sample Size Effects in Statistical Pattern Recognition: Recommendations for Practitioners , 1991, IEEE Trans. Pattern Anal. Mach. Intell..

[12]  John Owens,et al.  Streaming architectures and technology trends , 2005, SIGGRAPH Courses.