Python extension for a convolution neural network accelerator

Abstract

With the advent of artificial intelligence, machine learning, and deep learning, comes numerous possible possibilities for their applications. Neural networks that can self-learn is of particular importance. Convolutional Neural Networks (CNN) are one such neural network that has made significant strides. Image classification, recognition, object detection, and many other applications have seen a rise in our everyday lives, yet more can be done. Powering these networks through traditional hardware accelerators like our everyday graphical processing units (GPUs) prove powerful and efficient, but not portable. Field Programmable Gate Arrays (FPGA) are a type of hardware accelerator that has shown promise in delivering powerful computational capabilities that mesh well with the architecture of CNNs. To capitalise on their strengths, this report aims to document the seamless porting of architecture from a user-friendly Python environment directly to FPGAs for fast implementation.