preprint: paper on error-backpropagation in spiking neural networks

[Sander Bohte]

Apologies if you receive this posting more than once.

A preprint is available for download, of the paper 'SpikeProp:
Error-Backpropagation for Networks of Spiking Neurons', by Sander
Bohte, Joost Kok and Han La Poutré (to appear in `Neurocomputing').

For a network of spiking neurons that encodes information in the
timing of individual spike-times, we derive a supervised learning
rule, SpikeProp, akin to traditional error-backpropagation and show
how to overcome the discontinuities introduced by thresholding.  Using
this learning algorithm, we demonstrate how networks of spiking
neurons with biologically reasonable action potentials can perform
complex non-linear classification in fast temporal coding just as well
as rate-coded networks.  We perform experiments for the classical
XOR-problem, when posed in a temporal setting, as well as for a number
of other benchmark datasets. When comparing the (implicit) number of
biological neurons that would be required for the respective
encodings, it is empirically demonstrated that temporal coding
potentially requires significantly less neurons.  As we show that
reliable temporal computation can only be accomplished by
spike-response functions with a time constant longer than the coding
interval, the results also refute the long-standing argument against
temporal coding that states that the typical integration times of real
neurons are too long to compute a fast temporal code.

keywords: spiking neural networks, temporal coding,
error-backpropagation, XOR.

Download Instructions: Go to
http://www.cwi.nl/~sbohte/pub_spikeprop.htm and click on the file to
download (in PDF format [393K], or zipped
Postscript [515K]).

Comments Welcome

If you have problems downloading, please e-mail me.

Sander Bohte