Non-spiking implementations of spiking neurons and networks

In this work we explore the implementation of spiking neuron dynamics using non-spiking architectures. Using the wavelet analysis of spike trains as an inspiration, our approach is based on applying a transformation to the... [ view full abstract ]

In this work we explore the implementation of spiking neuron dynamics using non-spiking architectures. Using the wavelet analysis of spike trains as an inspiration, our approach is based on applying a transformation to the output of a neuron and then defining a conjugated network with neurons that both take as input and generate this transformed output. Based on the information retained, we can generate a series of conjugated networks with increasing degrees of complexity. For instance, at the zeroth order the resulting c-network reduces to a spike rate model, with successive order introducing the impact of other effects such as spike intensity, and synchronization.

One of the advantages of this approach is that it allows us to systematically study the dynamics of neural systems based on the amount and type of information transmitted between neurons. In this work we exemplify this approach using leaky integrate and fire neurons as a starting point.

From an implementation perspective, avoiding the need to implement spiking neurons while maintaining an equivalent functionality can greatly simplify the design of neuromorphic architectures. However, the validity of the approach presented here relies on single spikes and spike trains having different characteristic time scales.