TITLE: Learning to Map Spike Train Patterns in Multilayer Spiking Neural Networks
SPEAKER: Andre Gruning (University of Surrey)

ABSTRACT:
There is increasing evidence that the precise timing of spikes
generated by neurons, and not just their firing rate, conveys
meaningful information regarding input and output and internal processing of the nervous
system. It is also widely considered that synaptic plasticity forms
the basis of learning temporal spike-train-based codes in the
brain. In particular, spike-timing dependent plasticity is believed to
play a key role in learning, where changes in the synaptic strength
between paired neurons have a dependence on relative pre- and
postsynaptic spike times.

However, relating such localized plasticity changes to learning on the
network level remains a significant challenge. To address this, we
formulate a new supervised learning rule that can train spiking
networks containing a hidden layer of neurons to associate
spatio-temporal input and output spike patterns.

We demonstrate the high performance of the learning rule in terms of
the number of pattern associations that it can learn.  Our approach contributes both an understanding of how spike pattern information processing might take place in the brain,
and a learning rule that has technical potential for artificial neural networks build from spiking neurons