Building brains with arm processors and FPGAs based on high performance architectures convolutional neural networks

Lai, Jian Chang (2024) Building brains with arm processors and FPGAs based on high performance architectures convolutional neural networks. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full text) Building brains with arm processors and FPGAs based on high performance architectures convolutional neural networks.pdf - Submitted Version Download (2MB)

Abstract

The advent of Convolutional Neural Networks (CNNs) has transformed the landscape of artificial intelligence, particularly in visual information processing. This thesis embarks on a comprehensive exploration of advanced architectures for CNNs, focusing on the strategic integration of ARM processors and Field-Programmable Gate Arrays (FPGAs). The overarching goal is to harness the synergies between these heterogeneous computing platforms, capitalizing on their respective strengths to engineer high-performance systems capable of intricate visual interpretation. The research unfolds through an in-depth investigation into both hardware and software aspects, aiming to optimize the design, deployment, and performance of CNNs. Special attention is given to the development of tailored algorithms that align with the unique features of ARM processors and FPGAs. This includes the implementation of efficient memory utilization strategies and parallelization techniques to fully exploit the parallel processing capabilities inherent in these architectures. A critical facet of the study involves addressing challenges related to power consumption, thermal considerations, and resource utilization. By exploring novel approaches to mitigate these challenges, the thesis seeks to establish a foundation for creating intelligent systems with brain-like processing capabilities while maintaining energy efficiency. The research methodology employs rigorous experimentation and performance evaluations, with a keen focus on determining the trade-offs between computational efficiency and model accuracy. Insights derived from this exploration contribute to the overarching goal of advancing the field, offering a nuanced understanding of how the integration of ARM processors and FPGAs can yield optimized architectures for CNNs. The findings of this thesis not only extend the current understanding of high-performance architectures for CNNs but also lay the groundwork for future developments at the intersection of hardware design, neural networks, and artificial intelligence. The implications of this research resonate across various applications, from enhancing computer vision capabilities to empowering autonomous systems with sophisticated visual perception and interpretation.

Actions (login required)

View Item