Aristotle University Of Thessaloniki (AUTH)
Mapping efficient convolutional neural networks to resource-limited system-on-chip field-programmable gate arrays for mobile deep learning application acceleration
Abstract
dc:descriptionMobile deep learning applications running on consumer smartphones, drones, and autonomous vehicles have to satisfy the high computation requirement of the deep learning algorithms, and, at the same time, achieve low power consumption. Low parameter and high accuracy Convolutional Neural Networks (CNNs) have made a strong appearance in the deep learning research community. These efficient CNNs are the most suitable workload for mobile deep learning applications. Apart from using an efficient CNN, the required workload can be further reduced by using compression methods. This thesis presents the candidate CNN workloads, the methods used for applying compression on a CNN model, the development of a System-on-a-Chip (SoC) Field-Programmable Gate Array (FPGA) CNN accelerator template based on High-Level Synthesis (HLS) methods, an HLS-driven analytical performance model, a CNN-to-FPGA mapping workflow, and, finally, results, conclusions, and future work. Hopefully, this thesis and its results will be used for developing new research ideas in the field of mobile deep learning application acceleration using SoC FPGAs.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Mousouliotis, Panagiotis
- Μουσουλιώτης, Παναγιώτης
Subjects
dc:subject × 14- Επιτάχυνση αλγορίθμων
- Υψηλού επιπέδου σύνθεσης
- Συστήματα-σε-τσιπ με συστοιχία επιτόπια προγραμματιζόμενων πυλών
- Συνελικτικά νευρωνικά δίκτυα
- Algorithm acceleration
- High-level synthesis (HLS)
- System-on-chip field-programmable gate arrays (SoC FPGAs)
- Convolutional neural networks (CNNs)
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Υπολογιστές, Υλικό (hardware) και Αρχιτεκτονική
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Computer science, Hardware and Architecture
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/56389