Figure 5. Memory decay for a subset of potentiated synapses in a recurrent network with the in vitro parameter set.

(A) Temporal evolution of the average excitatory (red) and inhibitory (blue) firing rate. A network of 10,000 LIF neurons is initialised at the theoretically predicted steady-state and simulated for 20 min real time. (B) Temporal dynamics of synaptic efficacies in the network. The majority of synapses are initialised to the theoretically predicted asymptotic synaptic efficacy (mean: magenta; single synapse example: dark gray). A randomly selected subset of 5% are set to 1 at the beginning of the simulation (mean: green; single synapse example: light gray). (C) The exponential decay time constant of the potentiated synapses. The value obtained from fitting a single exponential to the mean decay (green dots) is well approximated by the analytically calculated decay time constant from the OU process (Eq. (3)). Introduction of a double-well potential does not modify the memory time constant for the in vitro parameter set (orange stars). The slight deviation of the decay time constants with respect to the OU theory, that is, the network decay time constants are slower, are due to the LIF firing statistics as can be seen from the comparison with independent LIF neurons (magenta dots).