Dynamical transitions in network models of collective computation

The field of neural network modelling has grown up on the premise that the massively parallel distributed processing and connectionist structure observed in the brain is the key behind its superior performance. The conventional network paradigm has mostly centered around a static approach-the dynamics involves gradient descent of the network state to stable fixed-points (or, static attractors) corresponding to some desired output. Neurobiological evidence however points to the dominance of non-equilibrium activity in the brain, which is a highly connected, nonlinear dynamical system. This has led to a growing interest in constructing nonequilibrium models of brain activity several of which show extremely interesting dynamical transitions, In this paper, we focus on models comprising elements which have exclusively excitatory or inhibitory synapses. These networks are capable of a wide range of dynamical behaviour, including high period oscillations and chaos. Both the intrinsic dynamics of such models and their possible role in information processing are examined