Using LeakyIntegrator with Projection

[David, Myungseok Shim]

Hi!

I have an issue using Leaky Integrator with Projection.

# Definition of the neuron

LIF = LeakyIntegrator(tau=20, B=-70, T=70, sum=('exc'))

# Projection

PIH = Projection(pre=groupi_in,post=group_hid,target='exc',synapse=RMSTDP_I)

PIH.connect_all_to_all(weights=Uniform(-5.0, 0.0))

When I compiled this, it showed the following error.

pop2.hpp:142:51: error: ‘exc’ was not declared in this scope

                 double _v =  __stepsize_v*(B[i] + exc - v[i]);

                                                   ^

In file included from ANNarchy.h:62:0,

                 from ANNarchyCore0.cpp:541:

proj0.hpp: In member function ‘void ProjStruct0::compute_psp()’:

proj0.hpp:159:26: error: ‘struct PopStruct1’ has no member named ‘g_exc’

                     pop1.g_exc[post_rank[i]] +=  w[i][j];

                          ^

I am not sure on the leaky integrator for implementing the following equation.

dVm/dt=(Vl-Vm)/tau : volt

Also, the error with projection maybe related with target. I think it's because of my inexperience with the library.

Regards,

Myung Seok 

[Julien Vitay]

Hi,

LeakyIntegrator is a rate-coded neuron, i.e. it has no spiking mechanism. You want to use IF_curr_exp or so in your case:

http://annarchy.readthedocs.io/en/stable/API/SpecificNeuron.html

Best

Julien

[David, Myungseok Shim]

Hi Julien.

Thanks for the reply. I attached a leaky integrate and fire model equation below, which I am trying to implement.

In Brian1 or 2, LIF is represented as following equation

lif_eqs = '''

dVm/dt=(Vl-Vm)/tau : volt

'''

# Hidden Neuron

hid = NeuronGroup(60, model=lif_eqs, threshold=-55*mV, reset=Vr)

# Synapse for providing spikes to target(Vm)

link_in_hid_in  = Connection(input, hid, 'Vm', weight=rand(len(groupi_in),len(group_hid))*-5*mV,sparseness=1)

In ANNarchy, I implemented LIF model using the example given by the documentation website.

LIF = Neuron(
	parameters="""
	tau = 20.0 : population
	Vl = -70.0 : population
	Vt = -54.0 : population
	Vr = -70.0 : population
	""",
	equations="""
	dv/dt = (Vl - v)/tau + g_exc : init=-70.0
	""",
	spike="""
	v > Vt
	""",
	reset="""
	v = Vr
	"""
	)

group_hid = Population(name='Hidden', geometry=60, neuron=LIF)

PIH = Projection(pre=groupi_in, post=group_hid, target='exc', synapse=)

Are those both same representation of LIF? 

Thanks,

Myung 

[Julien Vitay]

Hi,

yes, it sounds quite similar to the Brian implementation. Apparently you do not need a synapse type, and you can use PIH.connect_all_to_all(weights=Uniform(0.0, 5.0)) to create the projection.

Best

Julien 

[David, Myungseok Shim]

Hi Julien.

Thank you for the clarification.

Additionally, I tried the following equation and it worked close to the network based on Brian.

Now I am moving to implement Liquid State Machine based on ANNarchy. In aspect of the simulation speed, ANNarchy works at least 10 times faster than Brian2.

Thank you for bringing up the nice SNN package!

Regards,

Myung 

-------------------------------------

# Definition of the neuron

LIF = Neuron(

    parameters="""

        tau = 20.0 : population

        Vl = -70.0 : population

        Vt = -54.0 : population

        Vr = -70.0 : population

    """,

    equations="""

        dg_exc/dt = (Vl - g_exc)/tau : init=-70.0

    """,

    spike="""

        g_exc > Vt

    """,

    reset="""

        g_exc = Vr

    """

)