Skip to content

Emitters API

Neuron models and emitter implementations for neural dynamics simulation.

Izhikevich Model

jaxfne.set_emitter("izhikevich", preset="cortical_eig")

The Izhikevich neuron model is a phenomenological spiking neuron model with two state variables (v, u). It provides a good balance between computational efficiency and biological realism for tutorial-scale simulations.

State Variables

  • v (mV): Membrane potential (-90 to 30 mV typical range)
  • u (µA): Membrane recovery variable (adaptation/refractory effects)

Dynamics

dv/dt = 0.04*v² + 5*v + 140 - u + I
du/dt = a*(b*v - u)

If v ≥ 30 mV, spike occurs and states reset: 

v ← c
u ← u + d

Parameters

Parameters are organized in IzhikevichParams dataclass:

  • a (float): Recovery timescale (0.02 typical)
  • b (float): Coupling strength (0.2 typical)
  • c (float): Reset voltage (-65 mV typical)
  • d (float): Recovery reset (6 µA typical)
  • I_injected (float): Injected current baseline (optional)

Presets

Available presets define different neuron behaviors:

Preset Behavior Use Case
"cortical_eig" Regular spiking, moderate firing Tutorial default
"tonic_spiking" Continuous firing Sustained activity
"phasic_spiking" Burst then adapt Transient responses
"fast_spiking" High-frequency spikes Inhibitory interneurons
"chattering" Burst patterns Bursting neurons

Example: 

cfg = jtfne.Configuration()
cfg = cfg.set_emitter("izhikevich", "cortical_eig")

IzhikevichParams

jaxfne.IzhikevichParams

Parameter container for Izhikevich neuron model.

Attributes

  • a (float): Recovery timescale
  • b (float): Coupling strength
  • c (float): Reset voltage (mV)
  • d (float): Recovery reset (µA)

Methods

class_method(class_label: str) -> IzhikevichParams

Create parameters for a specific neuron class.

Parameters: - class_label (str): Neuron class (e.g., "excitatory", "inhibitory", "fast_spiking")

Returns: IzhikevichParams with preset values

Example: 

exc_params = jtfne.IzhikevichParams.class_method("excitatory")

ReceptorSpec

jaxfne.ReceptorSpec

Specification for synaptic receptor types and kinetics.

Attributes

  • name (str): Receptor name (e.g., "AMPA", "NMDA", "GABA_A")
  • tau_decay (float): Time constant (ms)
  • reversal (float): Reversal potential (mV)
  • conductance_scaling (float): Weight factor

Standard Receptors

Access via jaxfne.standard_receptor_specs():

  • AMPA: Fast excitatory (1–2 ms)
  • NMDA: Slow excitatory (100 ms)
  • GABA_A: Fast inhibitory (5–10 ms)
  • GABA_B: Slow inhibitory (100–200 ms)

Example: 

receptors = jtfne.standard_receptor_specs()

SynapseSpec

jaxfne.SynapseSpec

Specification for synaptic connections between neurons.

Attributes

  • source_idx (int): Source neuron index
  • target_idx (int): Target neuron index
  • receptor (str): Receptor type (e.g., "AMPA", "GABA_A")
  • weight (float): Synaptic strength (conductance)
  • delay (float, optional): Transmission delay (ms)

EIGNetwork

jaxfne.EIGNetwork

Excitatory-Inhibitory-Gap junction network representation.

Attributes

  • n_exc (int): Number of excitatory neurons
  • n_inh (int): Number of inhibitory neurons
  • synapse_specs (list[SynapseSpec]): Connection list
  • gap_specs (list[SynapseSpec], optional): Gap junction connections

Methods

to_dense() -> dict

Convert to dense adjacency matrix representation.

Returns: Dictionary with connectivity matrices

Example: 

network = jtfne.EIGNetwork(n_exc=100, n_inh=20)
dense = network.to_dense()

EdgeList

jaxfne.EdgeList

Sparse edge list representation of network connectivity.

Attributes

  • edges (jax.Array): Connection indices [2, num_edges]
  • weights (jax.Array): Synaptic weights [num_edges]

Methods

from_dense(adj_matrix: jax.Array) -> EdgeList

Create EdgeList from dense adjacency matrix.

Parameters: - adj_matrix (jax.Array): Dense connectivity matrix

Returns: EdgeList

Example: 

dense = jnp.ones((100, 100))
edges = jtfne.EdgeList.from_dense(dense)

Emitter Functions

simulate_eig_izhikevich(cfg, I_ext, seed) -> (spikes, V_m, u)

Simulate Excitatory-Inhibitory-Gap junction network with Izhikevich dynamics.

Parameters: - cfg (Configuration): Network configuration - I_ext (jax.Array): External input current [time, neurons] - seed (int): Random seed

Returns: Tuple of (spikes, membrane voltage, recovery variable)

Example: 

spikes, V_m, u = jtfne.simulate_eig_izhikevich(cfg, I_ext, seed=7)

make_eig_network(n_exc, n_inh, ...) -> EIGNetwork

Create an Excitatory-Inhibitory-Gap junction network structure.

Parameters: - n_exc (int): Number of excitatory neurons - n_inh (int): Number of inhibitory neurons - Additional connectivity parameters

Returns: EIGNetwork object

Example: 

network = jtfne.make_eig_network(n_exc=80, n_inh=20)

standard_receptor_specs() -> dict[str, ReceptorSpec]

Get standard synaptic receptor specifications.

Returns: Dictionary of receptor type → ReceptorSpec

Available types: AMPA, NMDA, GABA_A, GABA_B

Example: 

receptors = jtfne.standard_receptor_specs()
print(receptors["AMPA"])

Scope Notes

  • Izhikevich model is phenomenological: Not a detailed Hodgkin-Huxley model; suitable for tutorial and prototyping workflows
  • Spike threshold (v ≥ 30 mV): Fixed threshold; not voltage-dependent channels
  • Synaptic kinetics: Exponential decay; suitable for learning-scale simulations
  • Network connectivity: Declared via configuration; not learned (unless using optimization workflow)