Storage of Correlated Patterns in Standard and Bistable Purkinje Cell Models
Figure 2
Standard perceptron storing correlated input/output sequences.
A. Maximal capacity as a function of (dot: simulation, error bar: standard deviation, line: fit with a function
). Inset. Same but plotted as a function of
in a loglog scale. B. Maximal capacity as a function of the output correlation for different input correlations. C. Maximal capacity as a function of the input correlation for different output correlations. D. Weight distribution after learning at the maximal capacity for the case of uncorrelated input and output (blue: simulation, green: theory). The theoretical fraction of silent synapses is 0.5. The rest of the distribution is a truncated Gaussian with zero mean and standard deviation
. E. Fraction of silent synapses as a function of the output correlation for different input correlations. The theoretical value is 0.5 (green). F. Variance of the weight distribution normalized by the mean synaptic weight, fitted by a truncated Gaussian. The theoretical value is
(green). In all simulations, the perceptron has
inputs and the simulations were averaged over 10 trials. The coding level is
.