Massively Parallel kNN using CUDA on Spam-Classification
Joshua Smithrud, Patrick McElroy, Răzvan Andonie · 2015
Email Spam-classification is a fundamental, unseen element of everyday life. As email communication becomes more pro-lific, and email systems become more robust, it becomes in-creasingly necessary for Spam-classification systems to run accurately and efficiently while remaining all but invisible to the user. We propose a massively parallel implementation of Spam-classification using the k-Nearest Neighbors (kNN) al-gorithm on nVIDIA GPUs using CUDA. Being very simple and straightforward, the performance of the kNN search de-grades dramatically for large data sets, since the task is com-putationally intensive. By utilizing the benefits of GPUs and CUDA, we seek to overcome that cost. 1