Abstract
Associative processes in the brain may involve richer dynamics than displayed by fixed-point attractor networks. An input stimulus thus may generate an exploratory search of the space of stored patterns, corresponding to limit-cycle or chaotic dynamic behaviour, prior to the ultimate retrieval of a particular pattern in a fixed-point attractor. Of primary interest is a network design capable of an autonomous control of the required sequence of bifurcations. In this work an abstract neural network model of cortical associative memory is defined incorporating realistic features that determine the dynamic character and its transitions. The complexity of the network dynamics is controlled by the neuronal adaptability, i.e. the strength of coupling between activity and excitability of the network units. At weak adaptability the dynamics have fixed-point attractors and at strong adaptability more complex dynamics result, either limit cycles of varying complexity and period, or chaotic behaviour. In the brain neuronal adaptability in turn can be regulated by various neuromodulators. An autonomous regulation of the network dynamics can be based on an activity-dependent release of neuromodulators. For neurotransmitters serving both as modulators and in fast synaptic transmission, different types of receptor accomplish the different functions. The metabotropic receptors of glutamate and muscarinic receptors of acetylcholine are considered as examples of indirectly coupled receptors involved in modulatory mechanisms. Neuromodulators, such as the monoaminergic ones, released from axonal varicosities can also contribute to an autonomous behaviour. The study identifies certain features of biological systems that are crucial in determining neural network dynamics and function, thus it is also hoped that it will provide some guidance for further neurophysiological investigations. In artificial neural systems, dynamically controlled associative processes allow new types of applications.