Gradient Learning in Networks of Smoothly Spiking Neurons

0318366 - ÚI 2010 RIV DE eng C - Conference Paper (international conference)
Šíma, Jiří
Gradient Learning in Networks of Smoothly Spiking Neurons.
[Gradientní učení sítí hladce pulzních neuronů.]
Advances in Neuro-Information Processing. Revised Selected Papers Part II. Berlin: Springer, 2009 - (Köppen, M.; Kasabov, N.; Coghill, G.), s. 179-186. Lecture Notes in Computer Science, 5507. ISBN 978-3-642-03039-0.
[ICONIP 2008. International Conference on Neural Information Processing /15./. Auckland (NZ), 25.11.2008-28.11.2008]
R&D Projects: GA AV ČR 1ET100300517; GA MŠMT(CZ) 1M0545
Institutional research plan: CEZ:AV0Z10300504
Keywords : spiking neuron * back-propagation * SpikeProp * gradient learning
Subject RIV: IN - Informatics, Computer Science

A slightly simplified version of the Spike Response Model SRM0 of a spiking neuron is tailored to gradient learning. In particular, the evolution of spike trains along the weight and delay parameter trajectories is made perfectly smooth. For this model a back-propagation-like learning rule is derived which propagates the error also along the time axis. This approach overcomes the difficulties with the discontinuous-in-time nature of spiking neurons, which encounter previous gradient learning algorithms (e.g. SpikeProp). The new algorithm can naturally cope with multiple spikes and preliminary experiments confirm the smoothness of spike creation/deletion process.

Mírně zjednodušená verze modelu spiking (pulzního) neuronu SRM0 (Spike Response Model) je upravena pro gradientní učení. Konkrétně vývoj posloupností spiků podél trajektorií parametrů vah a zpoždění je dokonale hladký. Pro tento model je odvozeno
Permanent Link: http://hdl.handle.net/11104/0167808