Figure 4.
Target distributions learned with α-shaped response kernels. A, The target distribution P*(x) = Pw*(x) has the same parametrization as the model with excitatory weights from groups of 10 neurons projecting to subsequent groups of 10 neurons and inhibitory weights otherwise. B, Precisely timed spike patterns can be learned. C, The sequence of Figure 2A is encoded with periods of high or low firing rate. The target was generated by applying strong input current square pulses. Refractoriness sets an upper bound to the maximal firing rate. In recalls the model system was clamped to the target up to the green line.