Rapid application prototyping for hardware modular spiking neural network architectures

S Pande, F Morgan, F Krewer, J Harkin, LJ McDaid, B McGinley

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Spiking neural networks (SNNs) are well suitedfor functions such as data/pattern classification, estimation,prediction, signal processing and robotic control applications.Whereas the real-world embedded applications areoften multi-functional with orthogonal or contradictingfunctional requirements. The EMBRACE hardware modularSNN architecture has been previously reported as anembedded computing platform for complex real-worldapplications. The EMBRACE architecture employs geneticalgorithm (GA) for training the SNN which offers fasterprototyping of SNN applications, but exhibits a number oflimitations including poor scalability and search spaceexplosions for the evolution of large-scale, complex, realworldapplications. This paper investigates the limitationsof evolving real-world embedded applications withorthogonal functional goals on hardware SNN using GAbasedtraining. This paper presents a novel, fast and effi-cient application prototyping technique using theEMBRACE hardware modular SNN architecture and theGA-based evolution platform. Modular design and evolutionof a robotic navigational controller applicationdecomposed into obstacle avoidance controller and speedand direction manager application subtasks is presented.The proposed modular evolution technique successfullyintegrates the orthogonal functionalities of the applicationand helps to overcome contradicting application scenariosgracefully. Results illustrate that the modular evolution ofthe application reduces the SNN configuration search spaceand complexity for the GA-based SNN evolution, offeringrapid and successful prototyping of complex applications onthe hardware SNN platform. The paper presents validationresults of the evolved robotic application implemented onthe EMBRACE architecture prototyped on Xilinx Virtex-6FPGA interacting with the player-stage robotics simulator.
LanguageEnglish
JournalNeural Computing and Applications
Volume27
Issue number4
Early online date8 Feb 2016
DOIs
Publication statusE-pub ahead of print - 8 Feb 2016

Fingerprint

Network architecture
Computer hardware
Neural networks
Robotics
Hardware
Controllers
Collision avoidance
Pattern recognition
Scalability
Signal processing
Managers
Simulators

Keywords

  • Spiking neural networks
  • networks-on-chip
  • FPGAs
  • acceleration
  • ring topology

Cite this

@article{96073e0a28a64c48ae314ebe870e4723,
title = "Rapid application prototyping for hardware modular spiking neural network architectures",
abstract = "Spiking neural networks (SNNs) are well suitedfor functions such as data/pattern classification, estimation,prediction, signal processing and robotic control applications.Whereas the real-world embedded applications areoften multi-functional with orthogonal or contradictingfunctional requirements. The EMBRACE hardware modularSNN architecture has been previously reported as anembedded computing platform for complex real-worldapplications. The EMBRACE architecture employs geneticalgorithm (GA) for training the SNN which offers fasterprototyping of SNN applications, but exhibits a number oflimitations including poor scalability and search spaceexplosions for the evolution of large-scale, complex, realworldapplications. This paper investigates the limitationsof evolving real-world embedded applications withorthogonal functional goals on hardware SNN using GAbasedtraining. This paper presents a novel, fast and effi-cient application prototyping technique using theEMBRACE hardware modular SNN architecture and theGA-based evolution platform. Modular design and evolutionof a robotic navigational controller applicationdecomposed into obstacle avoidance controller and speedand direction manager application subtasks is presented.The proposed modular evolution technique successfullyintegrates the orthogonal functionalities of the applicationand helps to overcome contradicting application scenariosgracefully. Results illustrate that the modular evolution ofthe application reduces the SNN configuration search spaceand complexity for the GA-based SNN evolution, offeringrapid and successful prototyping of complex applications onthe hardware SNN platform. The paper presents validationresults of the evolved robotic application implemented onthe EMBRACE architecture prototyped on Xilinx Virtex-6FPGA interacting with the player-stage robotics simulator.",
keywords = "Spiking neural networks, networks-on-chip, FPGAs, acceleration, ring topology",
author = "S Pande and F Morgan and F Krewer and J Harkin and LJ McDaid and B McGinley",
year = "2016",
month = "2",
day = "8",
doi = "10.1007/s00521-015-2136-0",
language = "English",
volume = "27",
journal = "Neural Computing and Applications",
issn = "0941-0643",
number = "4",

}

Rapid application prototyping for hardware modular spiking neural network architectures. / Pande, S; Morgan, F; Krewer, F; Harkin, J; McDaid, LJ; McGinley, B.

In: Neural Computing and Applications, Vol. 27, No. 4, 08.02.2016.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Rapid application prototyping for hardware modular spiking neural network architectures

AU - Pande, S

AU - Morgan, F

AU - Krewer, F

AU - Harkin, J

AU - McDaid, LJ

AU - McGinley, B

PY - 2016/2/8

Y1 - 2016/2/8

N2 - Spiking neural networks (SNNs) are well suitedfor functions such as data/pattern classification, estimation,prediction, signal processing and robotic control applications.Whereas the real-world embedded applications areoften multi-functional with orthogonal or contradictingfunctional requirements. The EMBRACE hardware modularSNN architecture has been previously reported as anembedded computing platform for complex real-worldapplications. The EMBRACE architecture employs geneticalgorithm (GA) for training the SNN which offers fasterprototyping of SNN applications, but exhibits a number oflimitations including poor scalability and search spaceexplosions for the evolution of large-scale, complex, realworldapplications. This paper investigates the limitationsof evolving real-world embedded applications withorthogonal functional goals on hardware SNN using GAbasedtraining. This paper presents a novel, fast and effi-cient application prototyping technique using theEMBRACE hardware modular SNN architecture and theGA-based evolution platform. Modular design and evolutionof a robotic navigational controller applicationdecomposed into obstacle avoidance controller and speedand direction manager application subtasks is presented.The proposed modular evolution technique successfullyintegrates the orthogonal functionalities of the applicationand helps to overcome contradicting application scenariosgracefully. Results illustrate that the modular evolution ofthe application reduces the SNN configuration search spaceand complexity for the GA-based SNN evolution, offeringrapid and successful prototyping of complex applications onthe hardware SNN platform. The paper presents validationresults of the evolved robotic application implemented onthe EMBRACE architecture prototyped on Xilinx Virtex-6FPGA interacting with the player-stage robotics simulator.

AB - Spiking neural networks (SNNs) are well suitedfor functions such as data/pattern classification, estimation,prediction, signal processing and robotic control applications.Whereas the real-world embedded applications areoften multi-functional with orthogonal or contradictingfunctional requirements. The EMBRACE hardware modularSNN architecture has been previously reported as anembedded computing platform for complex real-worldapplications. The EMBRACE architecture employs geneticalgorithm (GA) for training the SNN which offers fasterprototyping of SNN applications, but exhibits a number oflimitations including poor scalability and search spaceexplosions for the evolution of large-scale, complex, realworldapplications. This paper investigates the limitationsof evolving real-world embedded applications withorthogonal functional goals on hardware SNN using GAbasedtraining. This paper presents a novel, fast and effi-cient application prototyping technique using theEMBRACE hardware modular SNN architecture and theGA-based evolution platform. Modular design and evolutionof a robotic navigational controller applicationdecomposed into obstacle avoidance controller and speedand direction manager application subtasks is presented.The proposed modular evolution technique successfullyintegrates the orthogonal functionalities of the applicationand helps to overcome contradicting application scenariosgracefully. Results illustrate that the modular evolution ofthe application reduces the SNN configuration search spaceand complexity for the GA-based SNN evolution, offeringrapid and successful prototyping of complex applications onthe hardware SNN platform. The paper presents validationresults of the evolved robotic application implemented onthe EMBRACE architecture prototyped on Xilinx Virtex-6FPGA interacting with the player-stage robotics simulator.

KW - Spiking neural networks

KW - networks-on-chip

KW - FPGAs

KW - acceleration

KW - ring topology

U2 - 10.1007/s00521-015-2136-0

DO - 10.1007/s00521-015-2136-0

M3 - Article

VL - 27

JO - Neural Computing and Applications

T2 - Neural Computing and Applications

JF - Neural Computing and Applications

SN - 0941-0643

SN - 1433-3058

IS - 4

ER -