Random activity at the microscopic neural level in cortex ("noise") sustains and is regulated by low-dimensional...

In this review I posit two levels of neural function. Microscopic activity is generated by neurons, to the extent that they act autonomously or in concert with networks of finite numbers of other neurons. Macroscopic activity is found in neuropil, where it depends on the sustained interaction of innumerable neurons. These levels coexist in cerebral cortex. Microscopic activity is manifested in the fraction of the variance of single neuron pulse trains (> 99.9%) that is both random and uncorrelated with pulse trains of other neurons in the neuropil. Macroscopic activity is revealed in the < 0.1% of the total variance of each neuron that is covariant with all other neurons in an area of neuropil comprising a population. It is best observed in dendritic potentials recorded as surface EEGs. The "spontaneous" background activity of neuropil at both levels arises from mutual excitation within a population of excitatory neurons. It is governed by a point attractor of the neuropil, which is actualized by the microscopic activity engendering the macroscopic state, and which acts as an order parameter to regulate the contributing neurons. The point attractor manifests a homogeneous field of white noise, into which sensory receptors send their microscopic stimuli. When neuropil comprises both excitatory and inhibitory neurons, the interactions at the macroscopic level lead to oscillations, manifesting a limit cycle attractor. When multiple areas of neuropil comprising a sensory system interact, then owing to their incommensurate characteristic frequencies and the long axonal delays between them, the system maintains a global chaotic attractor having multiple wings, one for each discriminable class of stimuli. Access to each wing is by stimulus-induced state transitions, leading to the construction of macroscopic chaotic patterns, that are carried to targets of cortical transmission by the microscopic noise. The chaotic modulation of the carrier noise is extracted by the targets through spatiotemporal integration, thereby retrieving the small covariance comprising the chaotic signal. Thus, controlled noise is the substrate for the meanings of stimuli that are expressed in chaotic patterns of sensory cortical activity.