Targeting specific neurons with external current using add_current_source in PyGeNN

[bakawulusai]

Hello!
If I have a neuron population with 5000 neurons. I need to apply Poisson-distributed input to specific neurons with the following indices:
[1196, 1231, 2793, 2680, 3892, 4378, 4775, 4906]
Does add_current_source support selective targeting of specific neurons within a population?

[neworderofjamie]

Hi there,

Like most things in GeNN, the solution is to create a customised model to meet your needs - in this case a current source model. Assuming you mean Poisson-distributed inputs like those provided by the built in PoissonExp model, the easiest thing to do is create a modified version of that model (copy-pasting from https://github.com/genn-team/genn/blob/master/include/genn/genn/currentSourceModels.h#L107-L128) where Init is a variable (can differ across neurons) rather than a parameter (the same across all neurons):

from pygenn import create_current_source_model

cs_model = create_current_source_model(
    "cs_model",
    params=["weight", "tauSyn", "rate"],
    derived_params=[("ExpDecay", lambda pars, dt: np.exp(-dt / pars["tauSyn"]),
                    ("ExpMinusLambda", lambda pars, dt: np.exp(-(pars["rate"] / 1000.0) * dt))],
    vars=[("Init", "scalar"), ("current", "scalar")],
    injection_code="""
    scalar p = 1.0;
    unsigned int numSpikes = 0;
    do
    {
        numSpikes++;
        p *= gennrand_uniform();
    } while (p > ExpMinusLambda);
   current += Init * (scalar)(numSpikes - 1);
   injectCurrent(current);
   current *= ExpDecay;
   """)

Then you can set up the Init variable per-neuron (the scaling factor was calculated in the derived parameter logic in the original model) and add the current source to the model:

poisson_params = {"weight": 0.1, "tauSyn": 0.5, "rate": 1000.0}

poisson_inject_inds = np.asarray([1196, 1231, 2793, 2680, 3892, 4378, 4775, 4906])
posson_init = np.zeros(num_neurons)
poisson_init[poisson_inject_inds] = poisson_params["weight"] * (1.0 - np.exp(-dt / poisson_params["tauSyn"])) * (poisson_params["tauSyn"]/ dt)
poisson_vars = {"current": 0.0, "Init": poisson_init}

model.add_current_source(pop_name + "_poisson", cs_model, neuron_pop, poisson_params, poisson_vars )

[bakawulusai]

Thanks for the reply！It's worked.