Cognition, Concurrency Theory and Reverberations in the Brain: in Search of a Calculus of Communicating (Recurrent) Neural Systems

We consider whether techniques from concurrency theory can be applied in the area of Cognitive Neuroscience. We focus on two potential applications. The first of these explores structural decomposition, which is effectively assumed by the localisation of function metaphor that so dominates current Cognitive Neuroscience. We take concurrency theory methods, especially Process Calculi, as canonical illustrations of system description notations that support structural decomposition and, in particular, encapsulation of behaviour. We argue that carrying these behavioural and notational properties over to the Cognitive Neuroscience setting is difficult, since neural networks (the modelling method of choice) are not naturally encapsulable. Our second application presents work on verifying stability properties of neural network learning algorithms using model checking. We thereby present evidence that a particular learning algorithm, the Generalised Recirculation algorithm, exhibits an especially severe form of instability, whereby it forgets what it has learnt, while continuing to be trained on the same pattern set.