Anthony R. White

Evolving axonal delay neural networks for robot control

This paper investigates the dynamical and control properties of a discrete spiking neural network model with axonal delays. After examining contemporary work on spike timing as a mechanism for neural coding, we introduce a simple axonal delay network model which, via coincidence detection, demonstrates the presence of biologically observed regimes such as sustained firing and the emergence of synchrony. We establish delay criteria allowing for the classification of three distinct regimes including global synchrony, complex firing, and dissipation. We then proceed to test this model in a robot light seeking task. Results show that evolving network delays is sufficient for solving the task. We conclude by hypothesizing that global synchronous firing is more suited to reactive behaviours while complex firing patterns may serve as an organizing mechanism for more indirect processing.

Dynamic memory for robot control via delay neural networks

We present a procedure to decode spatio-temporal spiking patterns in delay coincidence detection networks with stable limit cycles. We apply this to control a simulated e-puck robot to solve the t-maze memory task. This work shows that dynamic memory modules formed by coincidence detection neurones with transmission delays can be effectively coupled to produce adaptive behaviours.