Prof. Emo Welzl and Prof. Bernd Gärtner
|Mittagsseminar Talk Information|
Date and Time: Thursday, May 10, 2012, 12:15 pm
Duration: 30 minutes
Location: CAB G51
Speaker: Stephan Kollmann
Recently it has been shown that recurrent neural networks with initially random connections and weights can learn the topology of an external input using a rate based neuron model and a Hebbian learning rule. However it was not clear whether these results can be reproduced in a more biologically plausible setting. We show that similar results can also be achieved using a spiking neuron model and a STDP (Spike Timing Dependent Plasticity) learning rule that is based on triplets of spikes (it has been shown that experimental data can be explained well using such a rule). We further analyze the trained network's ability to exhibit certain soft-winner-take-all behavior, namely signal restoration, winner selection and cue integration.
Automatic MiSe System Software Version 1.4803M | admin login