ABSTRACT
Dynamical neural activity, which involves oscillatory and activity synchronization in neural firing, seems to have significant implications for understanding how the brain produces cognitive behavior. Computations based on dynamical activity require much more feedback information than is employed in traditional neural network approaches. The traditional neural network approach typically involves computation based on feedforward mathematical mapping from input to output. Several findings, including the limited correlational dimensionality of EEG activity, suggest that if dynamical neural activity contributes to information processing in the brain, than its processing capacity is very limited. I review the evidence suggesting that dynamical neural activity might be responsible for the limited processing capacity observed in voluntary/conscious processes in human cognition and propose the following hypothesis: the brain changes its information processing strategy from heavily relying on dynamical activity toward processing that relies on unidirectional mapping as learning and the development of skills increase (i.e. processing becomes more automatic). This transfer results in a decrease in the need for limited dynamically-based resources. Computations based on such a transfer might provide powerful information processing properties that optimize the complexity of the computational tasks in the brain.
