CHAOS IN THE CNS: THEORY AND PRACTICE
Walter J. Freeman
Chapter in: Ventriglia F (ed.) Neural Modeling and Neural Networks. New York, Pergamon Press. pp 185-216
Abstract
Behavior is enacted by massive numbers of neurons that continually transmit to and receive from each other. The activities of only a small fraction of them can actually be observed, but global activity patterns can be inferred by sampling, ensemble averaging, and modeling. Distinctions are made between time and space ensemble averaging to extract cortical input and output functions; between microscopic state variables of neurons and macroscopic state variables of cerebral cortex; and between input to cortex deriving from environmental stimuli and perceptual output constituting the meaning of stimuli for subjects. Coding of percepts is shown to occur by spatial patterns of amplitude modulation (AM) of common carrier oscillations that are self-organized by neuron populations. The AM patterns are modified during learning by changes in synaptic strengths to form nerve cell assemblies for each class of discriminated input. Each sensory cortex maintains a global chaotic attractor with multiple wings, one for each class. During learning the system bifurcates by an alteration of synaptic strengths. During an act of perception the system is driven from a basal state and confined in a wing that is selected by an activated nerve cell assembly in a process known as chaotic itinerancy. The state transition occurs by virtue of the amplitude-dependent nonlinear gain of the neurons comprising the assemblies. Readout of percepts is by spatial integration, which extracts the cooperative activity pattern and deletes extraneous sensory-driven details of cortical activity by smoothing. References are given to more detailed experimental and mathematical treatments of the topic.