Stochastic learning solves the stability-plasticity problem (Fusi et al., 2000a)

but raises new issues related to the generation of the proper noise driving

the synaptic dynamics. Here we show that a simple, fully deterministic,

spike-driven synaptic device can make use of the network generated vari-

ability in the neuronal activity to drive the required stochastic mechanism.

Randomness emerges naturally from the interaction of deterministic neu-

rons, and no extra source of noise is needed. Learning and forgetting

rates of the network can be easily controlled by changing the statistics of

the spike trains without changing any inherent parameter of the synaptic

dynamics.