Towards a rationale for neural stability: a model of neural computation and network architecture

Networks of Boolean processing cells with low connectivity are known to be inherently stable, in the sense that they exhibit only limited reverberatory activity among their possible state transitions. This paper discusses the value of such a network as a functional model of a neural system and, in the light of the observed decrease in stability with increasing cell connectivity, seeks to identify the features of network architecture and cell computation which act to protect network stability, thereby providing a framework for an understanding of neural stability.