Browse Prior Art Database

Cellular Neuronal Network Grids - a scientific solution with Grid computing

IP.com Disclosure Number: IPCOM000029871D
Original Publication Date: 2004-Jul-15
Included in the Prior Art Database: 2004-Jul-15
Document File: 4 page(s) / 90K

Publishing Venue

IBM

Abstract

A catenation between the universal paradigm of Cellular Neural Networks (CNN) and the innovative approach of grid computing in the on-demand area is given. CNN are a massive parallel solutions for solving non-linear problems, modelling complex phenomena in medicine, physics and data analysis as well as powerful image processing and recognition systems. They usually are simulated on local computer systems or build as dedicated VLSI-implementations. However, the research of complex CNN structures and settings require massive computing power and thus can benefit from multi-system open architectures which can be provided by the grid approach. Propositions of two different realizations with grid architecture in mind are given by introducing an algorithm of implementing such methods in a CNN software simulator.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 45% of the total text.

Page 1 of 4

Cellular Neuronal Network Grids

--- a scientific solution with Grid computing

Throughout this paper a catenation between the universal paradigm of Cellular Neural Networks (CNN) and the innovative approach of grid computing in the on-demand area is given. CNN are a massive parallel solution for solving non-linear problems, modelling complex phenomena in medicine, physics and data analysis as well as powerful image processing and recognition systems. They usually are simulated on local computer systems or build as dedicated VLSI -implementations. However, the research of complex CNN structures and settings require massive computing power and thus can benefit from multi -system open architectures which can be provided by the grid approach. Propositions of two different realizations with grid architecture in mind are given by introducing an algorithm of implementing such methods in a CNN software simulator. First a brief introduction to CNN is given. Afterwards, problems for the current determination of such networks are discussed and routes are shown, how this field of research can benefit from the opportunities, grid computing can deliver.

1. Cellular Neuronal Networks

    A CNN consists mainly of a local active complex - the cell, and a connection description to its neighbors. Cells are only locally coupled, typically only to its nearest neighbors in an n-dimensional alignment.

    Complex nonlinear couplings are realized through the output functions of each cell and the weight coefficients of the so called templates, which realize the connection weights between the cells. Those templates are usually but not necessarily the same for every cell.

    The dynamics of this kind of network can be described by state equations of the form

[This page contains 1 picture or other non-text object]

Page 2 of 4

+++-=)))))))x

dx

dt

r


i

(

(

t

t


r

r

(


t


z


r

(


t


A


B

r

(


y

r

vr r

(


t

r

(

u


i


j


i


j


j

j


j


j


S

S

i

i

(

)


r

r


)

r

(

where

  t
y
i i

r =

denotes the cell output function, A(y(t)), B(u(t)) the connection definitions, u(t) the input and z(t) a global bias.

2. The benefits of grid computing for CNN

    CNN and grid computing make use of the same ideas on different levels of abstraction and integration. On the one hand, first highly sophisticated CNN VLSI realizations exist, which are able to solve the CNN equations in real time with analog components. However, those chips are strongly limited by size and the nonlinearity they can carry. So, before designing such a highly sophisticated chip, the search for solutions and network structures needs to be performed with a CNN simulation system. The most common free software is SCNN [ www.uni-frankfurt.de/fb13/iap/e_ag_rt/SCNN/index.html], which is developed at University of Frankfurt (R. Kunz, G. Geis et al). So far, this software is limited to single thread which leads to long determination runs. In the last years, investigations were performed on how to speed up the search for new methods with this s...