| Description: |
Biological neurons are diverse. They do not only appear in different cell types, but also in different "dynamical types" that is different transitions between resting and spiking state. For example, class 1 neurons (with a continuous f-I curve) correspond to one dynamical type (SNIC) whereas class 2 neurons (with a jump in their f-I curve) belong to another (HOPF). The dynamical type is important because it determines encoding capabilities and can affect network dynamics like synchronization and bursting.
We would like to understand the variety of "dynamical types" currently accessible in artificial spiking networks (both simulated or on hardware) and discuss whether taking their properties into consideration could support neuromorphic computation. Among others, we would appreciate discussions around:
1) Which parameters in existing neurons in ASNN can change their dynamical type? For example, the LIF comes with one dynamical type, but a QIF can be switched via the reset parameter.
2) Are there training procedures for networks with neurons that have different dynamical types, including training of a parameter that can switch their type. This could allow us to investigate the distribution of dynamical types that emerges for different tasks.
3) Are there parameters in hardware implemented neurons that are difficult to control, but could switch the dynamical types? In biological neurons for example temperature can switch a neuron from a SNIC to a HOM type, which might underlie pathological seizures.
Related literature:
1) Insect asynchronous flight requires neural circuit de-synchronization by electrical synapses
Silvan Hürkey, Nelson Niemeyer, Jan-Hendrik Schleimer, Stefanie Ryglewski, Susanne Schreiber, Carsten Duch
bioRxiv 2022.02.02.478622; doi: https://doi.org/10.1101/2022.02.02.478622
Example, where neurons in a certain dynamical type (HOM) would naturally lead to a splay state as observed in the motor network |