Resetting Neurons and Synapses

[manuele09]

I am using the PyGeNN simulator (version 5) to implement a spiking neural network. My network includes four neuronal populations (pn, kc, mbon, and apln) and four synapse populations connecting them. I am encountering an issue when attempting to reset the network during simulation.
The reset functions I am using are based on the official PyGeNN tutorials:

#Reset Neurons Variables
def reset_neuron(pop, var_init):
    for var_name, var_val in var_init.items():
        pop.vars[var_name].view[:] = var_val
        pop.vars[var_name].push_to_device()

def reset_spike_times(pop):
    pop.spike_times.view[:] = -np.finfo(np.float32).max
    pop.spike_times.push_to_device()

#Reset Synapses
def reset_out_post(pop):
    pop.out_post.view[:] = 0.0
    pop.out_post.push_to_device()

The reset_spike_times() function sometimes fails. Specifically:

• When calling reset_spike_times(pn) or reset_spike_times(apln), I get an error because pop.spike_times is None.
• However, reset_spike_times(kc) and reset_spike_times(mbon) work without issues.
• If I switch the kc_mbon synapse from a custom weight update model (anti_hebbian) to a static synapse (StaticPulse), then pop.spike_times becomes None for all neuronal populations.

Network Configuration

Neuronal Populations

pn = model.add_neuron_population("pn", NUM_PN, "LIF", LIF_PARAMS, LIF_INIT)
kc = model.add_neuron_population("kc", NUM_KC, "LIF", LIF_PARAMS, LIF_INIT)
mbon = model.add_neuron_population("mbon", NUM_MBON, "LIF", LIF_PARAMS, LIF_INIT)
apln = model.add_neuron_population("apln", 1, if_model, IF_PARAMS, IF_INIT)

Synaptic Populations:

pn_kc = model.add_synapse_population("pn_kc", "SPARSE",
                                    pn, kc,
                                    init_weight_update("StaticPulseConstantWeight", {
                                        "g": PN_KC_WEIGHT}),
                                    init_postsynaptic(
                                        "ExpCurr", {"tau": PN_KC_TAU}),
                                    init_sparse_connectivity("FixedNumberPreWithReplacement", {"num": PN_KC_FAN_IN}))
kc_apln = model.add_synapse_population("kc_apln", "DENSE",
                                        kc, apln,
                                        init_weight_update(
                                            "StaticPulse", vars={"g": KC_APLN_WEIGHT}),
                                        init_postsynaptic("DeltaCurr"))
apln_kc = model.add_synapse_population("apln_kc", "DENSE",
                                        apln, kc,
                                        init_weight_update(
                                            "StaticPulse", vars={"g": APLN_KC_WEIGHT}),
                                        init_postsynaptic("ExpCurr", {"tau": 
kc_mbon = model.add_synapse_population("kc_mbon", "DENSE",
                                    kc, mbon,
                                    init_weight_update(
                                        anti_hebbian, KC_MBON_PARAMS, {"g": kc_mbon_g}),
                                    init_postsynaptic("ExpCurr", {"tau": KC_MBON_TAU}))

Custom Weight Update Rule (anti_hebbian)

anti_hebbian = create_weight_update_model(
    "anti_hebbian",
    params=["tau", "A", "Wmin", "Wmax"],
    vars=[("g", "scalar")],
    pre_spike_syn_code="""
        addToPost(g);
        const scalar dt = t - st_post;
        if (dt > 0) {
            const scalar timing = exp(-dt / tau);
            const scalar newWeight = g - (A * timing);
            g = fmax(Wmin, fmin(Wmax, newWeight));
        }
        """,
    post_spike_syn_code="""
        const scalar dt = t - st_pre;
        if (dt > 0) {
            const scalar timing = exp(-dt / tau);
            const scalar newWeight = g - (A * timing);
            g = fmax(Wmin, fmin(Wmax, newWeight));
        }
        """
)

Thank you for your help.

[neworderofjamie]

Thanks for your detailed post. Basically, spike times are only calculated when a weight update model requests them so, in your case, the anti_hebbian model on the kc_mbon synapse population is 'requesting' presynaptic spike times from the kc neuron population and postsynaptic spike times from the mbon neuron population. This is why reset_spike_times(kc) and reset_spike_times(mbon) are the only calls which work without error.

[manuele09]

Thanks for you answer.