polito.it
Politecnico di Torino (logo)

SpikExplorer: a tool for Design Space Exploration of Spiking Neural Network Architecture

Dario Padovano

SpikExplorer: a tool for Design Space Exploration of Spiking Neural Network Architecture.

Rel. Stefano Di Carlo, Alessandro Savino, Alessio Carpegna. Politecnico di Torino, Corso di laurea magistrale in Data Science And Engineering, 2024

[img]
Preview
PDF (Tesi_di_laurea) - Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives.

Download (6MB) | Preview
Abstract:

SpikExplorer: a tool for Design Space Exploration of Spiking Neural Network Architecture We live in the age of artificial intelligence, where high level abstractions of human brain consume enormous amount of energy to write an essay for high school kids. In this scenario it could be asked how if there exists a way to improve consumption keeping performance high, the answer is in our brain. The human neural network consumes infinitesimal amounts of energy compared to the stateof-the-art artificial neural networks such as GPT-4, in this way the structure and dynamics of human brain could became an inspiration in the development of pioneering models such as Spiking Neural Networks. In the last few years, Spiking Neural Networks (SNN) heavily spread through the machine learning community as a novelty approach to deep learning. Its basic unit, the spike, take inspiration from the human action potential transmitted by neuron in neuron. These architectures have already found application in popular task as image classification and speech recognition thanks to the intrinsic relationship between spikes and time. The aim of this work is not only the use of SNN as a way to exploit the spiking dynamic similar to what happens between biological neurons, instead the goal is the creation of a tool capable of generalize as much as possible the construction of the SNN architecture by specifying particular constraints related to the future deployment on FPGA-based (Field Programmable Gate Array) neuromorphic hardware. This work is divided in two parts, the first will address the problem of finding the optimal method to search the design space, while the seconds will explore the tool structure with an extensive search carried on in order to evaluate the performance of the tool.

Relatori: Stefano Di Carlo, Alessandro Savino, Alessio Carpegna
Anno accademico: 2023/24
Tipo di pubblicazione: Elettronica
Numero di pagine: 98
Soggetti:
Corso di laurea: Corso di laurea magistrale in Data Science And Engineering
Classe di laurea: Nuovo ordinamento > Laurea magistrale > LM-32 - INGEGNERIA INFORMATICA
Aziende collaboratrici: Politecnico di Torino
URI: http://webthesis.biblio.polito.it/id/eprint/30981
Modifica (riservato agli operatori) Modifica (riservato agli operatori)