PYTHON AND CUDA FOR PARALLEL PROCESSING: AN UNDERGRADUATE CASE STUDY ACCELERATING NEURAL NETWORK MAMMOGRAM PROCESSING

Abstract

Originally confined to computer graphics, theGPU has emerged as a significant application accelerator inthe parallel programming arena. NVIDIA, with the CUDAprogramming framework, has made a substantialcontribution to making the GPU facilities accessible. Inaddition, the PyCUDA project provides Python languagebindings to the CUDA infrastructure. This paper profilesthe role of Python and CUDA in a parallel processingcourse by featuring an undergraduate project implementinga convolutional filter to accelerate neural networkprocessing of large mammogram images as a case study.Specifically, we describe the course, mammogram imagesand processing, the Python and CUDA interfaces, and theresulting speedup, along with observations on pedagogy.Index Terms ⎯ Parallel processing, CUDA, Python,pedagogy.