Neural networks and chaos

Describes current investigation of the potential uses of complex dynamical behavior for information processing in neural networks. The main implication is that apparently chaotic fluctuations of activity in biological neural networks should not be ignored by considering average firing-rates. It is shown that fluctuations of firing rates in neural networks can either be used in information processing or can be indicative of deeper, even more complex forms of information processing. The first model considers unstable nonequilibrium fluctuations in the average firing rate. This model demonstrates that dynamical behavior can be used to construct neural networks with a kind of multivalued logic or membership function as in fuzzy sets. The second model considers the complex dynamics at the level of individual action potential pulses