www.neuron.yale.edu

I wish to implement a intfire model with excitatory and inhibitory currents each having tau = 0.33 [ms] and the leaky membrane having taum = 10 [ms]. IntFire2 will not allow this due to the requirement of taus > taum. I am therefore investigating implementation using IntFire4. IntFire4 however has time constants for two inhibitory contributions, if I understand correctly. Thus I have been trying to eliminate the effect of one of the inhibitory currents. I have tried doing so by implementing the following cell definition:
However, my simulations are taking much more computational resources than when I had been using IntFire1 models (with refractory period of 5[ms]). My experience in computational neuroscience is limited compared to programming experience, and therefore I am wondering if I have made a mistake in my assumptions when trying to eliminate i1 from the model above. Alternatively, is IntFire4 simply much more demanding on resources compared to IntFire1 (potentially related to the refractory period)?

Thank you.

tseredynski wrote:I wish to implement a intfire model with excitatory and inhibitory currents each having tau = 0.33 [ms] and the leaky membrane having taum = 10 [ms].

Why not use IntFire1? It will do exactly what you describe. An input with weight > 0 will be excitatory, and with weight < 0 will be inhibitory.

IntFire4 however has time constants for two inhibitory contributions

Not so. IntFire4 has two "synaptic currents"--one excitatory, with time course governed by a single time constant taue, and the other inhibitory, time course governed by two time constants taui1 and taui2 (you may think of these as affecting principally the rising and falling phases of the inhibitory current).

Hm, i must be missing something obvious. I was under the impression IntFire1 only allowed specification of a refractory period and the leaky membrane time constant. Is there a way to specify the time constant associated with excitatory/inhibitory current as well? (or is this somehow implied in the other two variables?) I would prefer to use IntFire1 as there is also a refractory period I'd like to specify in my model, and had planned on needing to add this to IntFire4.

In reply to the discussion of inhibitory current in intfire4: thank you, I had forgotten that the two time constants affect the same current.

tseredynski wrote:I was under the impression IntFire1 only allowed specification of a refractory period and the leaky membrane time constant. Is there a way to specify the time constant associated with excitatory/inhibitory current as well?

The conceptual model embodied in IntFire1 is that the duration of synaptic current is much shorter than the membrane time constant. If you want excitation and inhibition to have different time courses, IntFire4 will do the job.

If you want a mechanism that offers both a refractory period and different time courses for excitation and inhibition, it will be necessary to implement a new class of artificial spiking cell. This could be based on IntFire4, but the question is how refractoriness is to be represented. The minimal representation would be for the model cell's membrane state to be forced to 0 for the duration of refractoriness, while the synaptic currents are allowed to evolve in time and respond to new input events. Would this be satisfactory for your purposes?

ted wrote: The conceptual model embodied in IntFire1 is that the duration of synaptic current is much shorter than the membrane time constant. If you want excitation and inhibition to have different time courses, IntFire4 will do the job.

Ah, so if I wished to approximate the synaptic current time constant as 0 compared to the membrane time constant than I could simplify my model to intfire1.

ted wrote: If you want a mechanism that offers both a refractory period and different time courses for excitation and inhibition, it will be necessary to implement a new class of artificial spiking cell. This could be based on IntFire4, but the question is how refractoriness is to be represented. The minimal representation would be for the model cell's membrane state to be forced to 0 for the duration of refractoriness, while the synaptic currents are allowed to evolve in time and respond to new input events. Would this be satisfactory for your purposes?

This would work for what I was thinking, however I will consider using intfire1 instead for sake of simplicity.

Thank you very much.