Neuron and Synapse Discussion Group Report

'Paul Hasler', Tara Julia Hamilton

This year we had one meeting of the neuron and synapse discussion group. We covered a number of topics. Namely, the need (or lack of need) for elaborate neurons in spiking neural networks, the “killer ap” i.e. what can spiking neural networks do better than other networks and the future of hardware neural networks. The “killer ap” inspired the most animated conversation. The general feeling is that the killer application for spiking networks is yet to be found. A number of people believed that the advantage in spiking networks is power consumption although this is obviously not the case if rate coding is simply used. In order to prove the utility of spikes a member of the group, Daniel Lofaro, created a simulation showing the reduction in power consumption when using timing codes rather than rate codes and that a useful compromise between speed and power could be found when both rate and timing codes were combined (see Figure 1). This line of research will continue as it provides valuable quantitative comparisons between communications protocols.

Figure 1. A comparison between power usage in rate code communications, time code communications and a combination of them both.

During the discussion another member of the group, Ryad …, suggested that the best way to show the utility of spiking networks is to model the behaviour of the human retina precisely. He gave a convincing argument as to why this may result in showing promising results for a spiking network. There was a consensus that we still do not have enough information from neuroscientists in order to realise the best abstraction of our neuron and synapse models. There was also a consensus that the reason for continuing research in this area was two-fold: to assist in the understanding of computational neuroscience AND to develop engineering tools which benefit from the efficiency of biology. The later reason requires further research into discovering the best way to use spikes.