Background firing in LIF model

Science, Projects, and Showcases
,
1

Hello everyone,

Is there any way to add background firing as random white noise to the LIF neurons?
I added the noise term in my equation, but the results do not look like a background firing activity; it seems it just added noise to the membrane potential trajectory.

eqs= '''
        dv/dt      = (-(v-E_l) + I_exc + I_inh)/tau_m + sqrt(2* sigma**2/tau_m)*xi: volt (unless refractory)
        dI_inh/dt = -I_inh/tau_i   : volt 
        dI_exc/dt = -I_exc/tau_e  : volt
'''
2

Hi @elnaz91. Can you give some more details of what you mean by “background firing as random white noise”? How would this look different from added noise to the membrane potential trajectory?

3

Hi @mstimberg,If you look at the plots below, the right one is Poisson input, to which I easily added the background rate to my input firing (after 100ms we have only background firing). For the left plot, although I added the noise term in my equation, it does not show any background firing. Note that the maximum weight values are in an order of e-3, and I do not want to change that. That’s why you see a very sparse firing rate, and I want to fix that with background firing.

image
image1238Ă—498 47.3 KB

4

Hi @elnaz91. If you want to have your neurons fire without input, then you either need to add a constant background current, noise, or both. Adding noise alone (as in your equation) will not trigger spikes unless you make it very strong. On the other hand, you probably do not want to make it that strong, because it would mean that the noise completely dominates the membrane potential, and the synaptic inputs will not have much effect. Since the “noise” is usually meant to be a model of many incoming spikes from outside the network (the part that is not modelled explicitly as your input), a better way is to use either a PoissonInput object, or a constant background current + noise. To get irregular firing, you should choose the background current to be sub-threshold on its own, i.e. so that it needs the fluctuating noise to cross the threshold. For the right parameters PoissonInput and constant current + noise will give very similar results.

1 Like
5

Oh, Thank you so much. That makes a lot more sense. I used PoissonInput + noise, and it worked for me.