Figure 1

The stabilization of old (unused) memory states by the combination of unstructured noise and antisymmetric STDP learning rule. (A) STDP learning rule considered. (B) The rate of change of the contribution of an unused state ($d{c}^{\mathsf{\text{unused}}}/dt$) to the synaptic weight matrix as a function of the contribution itself ($c^{\mathsf{\text{unused}}}$). The contribution in this case has two stable points, near zero and at a finite value. The former/latter stable points correspond to the unused memory pattern being absent/present in the network connectivity. At the stable points the rate of change of the pattern's contribution is zero. Small perturbations from the stable point will induce the rate of change that returns the system back to the stable point. The third point where the rate of change is zero is unstable and is therefore called the transition point.