LiveSNN: a new ecosystem for HEENS architecture

This project proposal and development of a several tools, a communication protocol and an embedded program for giving a neural network called HEENS that it is currently developed by the group ISSET from UPC the capability of being controlled remotely, and to extract the neural . This will provide a...

Descripción completa

Detalles Bibliográficos
Autor: Oltra Oltra, Josep Angel|||0000-0002-9706-1736
Tipo de recurso: tesis de maestría
Fecha de publicación:2020
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/334928
Acceso en línea:https://hdl.handle.net/2117/334928
Access Level:acceso abierto
Palabra clave:Neural networks (Computer science)
Embedded computer systems
SNN
Neural Network
HEENS
Assembler
neural synthesis
embedded
Xarxes neuronals (Informàtica)
Sistemes incrustats (Informàtica)
Àrees temàtiques de la UPC::Enginyeria electrònica
Descripción
Sumario:This project proposal and development of a several tools, a communication protocol and an embedded program for giving a neural network called HEENS that it is currently developed by the group ISSET from UPC the capability of being controlled remotely, and to extract the neural . This will provide a faster development, user friendly tools for the development and analysis of spiking neural networks for emulation and verification of biological neural networks and neural models. As such, three new pieces of software are developed called: LiveSNN protocol, which communicates the HEENS architecture to a remote PC for Supervisory Control And Data Acquisition (SCADA) operations, LiveSNN program, which manages the conections and supervises the activity of the HEENS, and HEENS Toolchain Suite (HTS), which is an upgrade of the previous synthesis and assembler tools in order to allow the implementation of more sophisticated neural networks being easy and be capable of optimise the assembler model of the neuron for the architecture. With those developments, the time to generate spiking neural networks, to debug them, and emulate them is increased in addition to the reduction of possible human errors due to the increased automation workflow that those programs give to the end user.