Neural networks and the representation of time

Abstract

The neurons within artificial neural networks have rather limited temporal behaviour, and the most common mechanisms used to improve their representation of temporal events are simple delay lines or recurrent feedback connections. While there are problems in the match between these properties and the neurophysiological realities, many of these problems are overcome by the benefits of multi-neuron networks. A simple network model is described, more as an example of the network approach than as a serious model of timing. However, it demonstrates that even the most crude network of neurons could be used to count or to accumulate the output of an internal clock.