SpikingNeuralNetworks.jl Documentation

Julia Spiking Neural Networks (JuliaSNN) is a library for simulating biophysical neuronal network models.

This documentation is work in progress; please contact me via the GitHub repository if you have any specific questions or want to collaborate!

Simple and powerful simulation framework

The library's strength points are:

• Modular, intuitive, and quick instantiation of complex biophysical models;
• Large pool of standard models already available and easy implementation of custom new models;
• High performance and native multi-threading support, laptop and cluster-friendly;
• Access to all models' variables at runtime and save-load-rerun of arbitrarily complex networks;
• Growing ecosystem for stimulation protocols, network analysis, and visualization (SNNUtils, SNNPlots, SNNGeometry).

SpikingNeuralNetworks.jl leverages the JuliaSNN ecosystem, which offers SNNPlots to plot models' recordings and SNNUtils for further stimulation protocols and analysis.

Models: populations, connections, and stimuli

SpikingNeuralNetworks.jl builds on the idea that a neural network is composed of three classes of objects: the network populations, their recurrent connections, and the external stimuli they receive. Thus, a SNN model is simply a NamedTuple with keys: pop, syn, stim. The element associated with the keys must be concrete subtypes of AbstractPopulation, AbstractConnection, or AbstractStimulus.

Models can be generated using compose with any population, connection, or stimulus type as keyworded arguments. The user can define the model by associating the correct subtypes to the named tuple, but we advise against it. For example:

using SpikingNeuralNetworks

E = SNN.IF(N = 100) # create an Integrate-and-Fire model population with 100 neurons. Use default parameters

EE = SNN.SpikingSynapse(E, E, :ge, w = rand(E.N, E.N)) # connect the populations with recurrent, spiking synapses, targeting the :ge field.
my_model = SNN.compose(E=E, EE=EE) # create a model with the E population and the EE connection.
# my_model = SNN.compose(;E, EE) # equivalent

compose assigns the correct types to the pop and syn and carries further integrity checks. The population and synapse elements will be assigned to my_model.pop.E and my_model.syn.EE, respectively.

Pre-existing models

For each subtype, JuliaSNN offers a library of pre-existing models. In the case above, an integrate-and-fire population (IF<:AbstractPopulation), a spiking synapse (SpikingSynapse<:AbstractSynapse). The collection of available models can be found under Models.

Models can also be extended by importing the AbstractPopulation, AbstractConnection, or AbstractStimulus types. Guidelines on how to create new models are presented in Models Extensions (WIP)

Simulation

Leveraging Julia's multiple dispatch, the simulation loop calls the methods defined for each type of model and parameter:

function sim!(...)
    update_time!(T, dt)
    for s in stimuli
        s_type = getfield(s, :param)
        stimulate!(s, s_type, T, dt)
        record!(s, T)
    end
    for p in populations
        p_type = getfield(t, :param)
        integrate!(p, p_type, dt)
        record!(p, T)
    end
    for c in connections
        c_type = getfield(c, :param)
        forward!(c, c_type)
        ## if train!(...) 
            plasticity!(c, c.param, dt, T)
        record!(c, T)
    end
end

In a loop step, the first to be activated are the stimuli which provide inputs to the populations. Thus, the differential equations associated to the populations are integrated. Finally, the population activity is propagated through the synapses (connections!).

Using Julia's passing-by-sharing, connections and stimuli maintain internal pointers to the populations' fields they are attached to. This allow to seamlessy read and updates the populations variables within the stimulate! and forward! functions.

Installation

JuliaSNN/SpikingNeuralNetworks.jl is not yet available on the public Julia repository! For the moment ]add SpikingNeuralNetworks will still direct you to the old version of the package.

You can install the latest version directly from the git repository:

]add https://github.com/JuliaSNN/SpikingNeuralNetworks.jl

To learn how to use the library you can follow the Tutorial.