Neural Coding via Transmission Delay Coincidence Detectors: An Embodied Approach

This thesis aims to contribute to the field of cognition via a careful investigation at the mesoscopic level of neural organization and activity via spatiotemporal dynamics to subserve reactive and adaptive behaviour. In particular, coding via coincidence detection with prorogation delays is adopted as the primary mechanism to investigate complex neural dynamics. Initially hypothesized by Moshe Abeles in the early 1980’s, coincidence detection enables neurones to respond by emitting an action potential (a spike) only when other input neurones spike in a precise order. In contrast to the traditional interpretation of neural behaviour whereby cells integrate inputs over long periods of time, coincidence detectors are sensitive to changes at millisecond or sub-millisecond time scales. However, networks of coincidence detectors that incorporate propagation delays remain poorly understood with respect to dynamics and functional application for cognitive tasks. After introducing the functional and biological evidence that motived this research, we explore existing work related to spatiotemporal coding and the neurone and network parameters that influence their behaviour. We then present a discrete neural network model that is used to expose the relationship between structural parameters and network dynamics. After applying these to a reactive light-seeking robot task, we find network parameters that enable dynamic memory storage of input spike patterns. Making use of these dynamics, we then introduce a method for decoding these memories in a modular network architecture and test this approach for robot control in memory maze tasks. Furthermore, the limits of the potentially high memory capacity of these networks is then tested by empirically evaluating both the noise tolerance and the memory capacity of these networks. To complement this empirical work, we develop a set of formal expressions which attempt to approximate analytically the amount of activity of these networks and the probability for any given spike pattern to be expressed by them.