AI summary

This paper introduces high-performance pattern recognition algorithms using a commodity Graphics Processing Unit (GPU), achieving over 300x speedup for Parzen windows and 100x speedup for ANN, using NVIDIA 8800 GTX GPU on NIST data set. These methods are compared to CPU implementations.

Abstract

The pattern recognition (PR) process uses a large number of labelled patterns and compute intensive algorithms. Several components of a PR process are compute and data intensive. Some algorithms compute the parameters required for classification directly for each test pattern using a large training set. Most algorithms have a training step, the results of which are used by a computationally cheap classification step. In this paper, we present high-performance pattern recognition algorithms using a commodity Graphics Processing Unit (GPU). Our algorithms exploit the high-performance SIMD architecture of GPU. We specifically study the Parzen windows scheme for density estimation and the Artificial Neural Network (ANN) scheme for training and classification in this paper. We present fast implementations of these on a NVIDIA 8800 GTX GPU. Our implementation of Parzen windows can simultaneously estimate probability values for 1K test patterns in about 14ms based on an input data set of 16K patterns. Our ANN can run an epoch of batch-training on the NIST data set with 56K 484-dimensional patterns and 10 output categories in less than 200 milliseconds. The speedup is more than 300 times for Parzen windows and 100 times for ANN over the CPU implementations using a commodity GPU that costs about $400.