Neurons can be simulated using different models that vary in the degree of biophysical realism. When designing an artificial neuronal network, I am interested in the consequences of choosing a degree of neuronal realism.
In terms of computational performance, the FLOPS vary from integrate-and-fire to the Hodgkin–Huxley model (Izhikevich, 2004). However, properties, such as refraction, also vary with the choice of neuron.
When selecting a neuronal model, what are consequences for the ANN other than performance? For example, would there be trade-offs in terms of stability/plasticity?
Izhikevich investigated the performance question in 2004. What are the current benchmarks (other measures, new models)?
How does selecting a neuron have consequences for scalability in terms of hardware for a deep learning network?
When is the McCulloch-Pitts neuron inappropriate?
Izhikevich, E. M. (2004). Which model to use for cortical spiking neurons? IEEE Transactions on Neural Networks, 15(5). https://www.izhikevich.org/publications/whichmod.pdf