Helene Paugam-Moisy, Regis Martinez and Samy Bengio (2007)
"A supervised learning approach based on STDP and polychronization in spiking neuron networks."
Proc. of European Symp. On Artificial Neural Networks - Advances in Computational Intelligence and Learning, pp. 427--432
==================================================================================================
Abstract
... We propose a network model of spiking neurons, without preimposed topology and driven by STDP,
a temporal Hebbian unsupervised learning mode, biologically observed. The model is further driven by a supervised learning algorithm, based on a margin criterion, that has effect on the synaptic delays linking the network to the output neurons, with classification as a goal task.




We use excitatory and inhibitory temporal windows as proposed in [7] and apply a multiplicative weight update.

[7] D. Meunier and H. Paugam-Moisy (2005)
"Evolutionary supervision of a dynamical neural network allows learning with on-going weights."
In IJCNNf2005, Int. Joint Conf. on Neural Networks, pp. 1493-1498.



The delays dij take integer values, randomly chosen in {1, . . . , 20}, both in the internal network and towards output neurons. ... The goal of the supervised learning mechanism we propose is to modify the delays from active internal neurons to output neurons


After an initialization phase generating a high disordered activity,
a learning phase is run, from 2000 to 11000ms (in simulated biological
time), with successive alternated presentations of two input patterns, similar
to Izhikevich stimulation patterns



Second, we analysed the evolution of the polychronous groups activation at two stages of the learning phase (Figure 4).