Redes neurais convolucionais implementadas em Field Programmable Gate Array (FPGA) e controlado por Paradigma Orientado a Notificações (PON)

Abstract

Artificial Intelligence is an upcoming paradigm that recently has made rapid advances, creating new aplications and use cases in several areas such as Computational Vision, Audio Processing, Data Science and Natural Language Processing, achieving results that are equal or better than human capability and/or algorithms manually designed by humans. One of the techniques which has seen wide spread adoption is that of Convolutional Neural Network, which has many applications in image and audio processing, objection detecting and even in self-driving cars. Due to this adoption, there is a new need of the efficient execution of those networks, both in speed and in energy efficiency, with one of the potential alternative to the traditional paradigms utilized by such networks being the Notification Oriented Paradigm which has advantages in parallel processing due to it’s cohesive and decoupled nature. In a similar manner, as a platform Field Programmable Gate Array(FPGA) demonstrate potential due to it’s reconfigurability at the hardware level and the possible paralelism, which is synergestic with NOP. This works implements an architecture as a proof of concept, dedicated to Convolutional Neural Networks in FPGA, using the Notification Oriented Paradigm in a hybrid model.