Parallel Distributed Infrastructure for Minimization of Energy

Biologically Sound Neural Networks for Embedded Systems Using OpenCL

Publication Type:

Conference Paper


NETYS 2013, Springer LNCS Vol. 7853 2013, Marrakech, Morocco (2013)




Neural Networks, Parallel Programming, Resource Awareness


In this paper, we present an OpenCL implementation of a biologically

sound spiking neural network with two goals in mind: First, applied neural dynamics
should be accurate enough for bio-inspired training methods, thus resulting
network data is reproducible in "in vitro" experiments. The second is that the implementation
produces code that runs adequately on up-to-date embedded graphical
chips for fast on-board classification applications, e.g., video image processing.We
describe the necessary steps required to implement an efficient algorithm using the
OpenCL framework and present evaluation results of the execution time compared
to traditional serial CPU code. We show that an optimized GPU kernel code can
perform sufficiently fast to be used for future embedded neural processing.



This paper was written in collaboration with the NES institute at Alpen-Adria-Universität Klagenfurt, Austria.

Programming for embedded systems is especially challenging, because one has to consider limited resources (e.g., battery life) and has therefore to be specifically efficient. Solutions for embedded systems and the resulting ideas can be borrowed for energy-efficient computing in the larger scale as well.

The implemented neural network and its performance improvements were part of the real-world use case discussion in ParaDIME