Metastable associative network models of dream sleep

Up to the present day, simulations using a neural network model have been carried out under the global inhibition and the uncorrelated noise in order to simulate the dynamics transition of neuronal activities from the flat (slow wave sleep) to the 1/f (dream sleep) power spectral density profile during the sleep cycle in cats. In particular, the metastability of the network attractor is shown to be essential in generating the 1/f fluctuations. Here, the dynamics of neuronal and network activities are analyzed under the correlated noises mimicking a cholinergic drive. Regardless of the network structure, symmetry and asymmetry, the behavior of network activity and the escape time distributions show that the long-lasting autocorrelation of the noise prolongs its prescence in the metastable states. These results and the newly estimated network attractor show that the interplay between the metastability of the network attractor and the noise statistics determines the dynamics of network activity. Our results may be able to provide the novel framework to investigate the function of dreaming in the higher-order brain function.