ΣΧΕΔΙΑΣΜΟΣ ΝΕΥΡΩΝΙΚΩΝ ΔΙΚΤΥΩΝ ΣΕ FPGA

Abstract

Abstract Artificial Intelligence has become an indistinguishable part of our daily life. It is undeniable that “smart algorithms” behind search engines and programs simplify and ease, in a tremendous level, the requested tasks and achieve at the same time extremely good results (incomparable to those of people) without the huge effort needed by human. As the domain of Artificial Intelligence evolves over the years, it contains a big variety of tools and methods in order to solve a problem. Artificial neural networks constitute a significant part of machine earning domain, which itself is a huge chapter of Artificial Intelligence. This dissertation focuses on the design and development of an artificial neural network in FPGA in order to solve a regression problem. Moreover, in this dissertation an in-depth analysis of artificial neural network (perceptron, spiking, etc.) characteristics is done. The design and implementation of the neural network is realized with MATLAB and Vivado software. Finally, the expected behavior of the neural network in comparison with the one running in the computer is examined and evaluated.