Build an event-driven E/I network

brainevent provides the event-driven communication layer of a spiking network: spikes fan out through a sparse connectivity structure to deliver synaptic currents. This guide shows the core pattern, then points to complete, runnable COBA and CUBA implementations of the classic Vogels–Abbott balanced network.

The core pattern

A population emits a binary spike vector each step; multiplying it by a connectivity structure delivers currents to the post-synaptic population. Splitting the population into excitatory and inhibitory groups with opposite-sign weights gives a balanced network:

import brainevent
import brainunit as u
import jax.numpy as jnp

n_exc, n_inh = 3200, 800
num = n_exc + n_inh

# Fixed fan-out connectivity, excitatory (+) and inhibitory (-)
exc_conn = brainevent.FixedPostNumConn(num_pre=n_exc, num_post=num,
                                       conn_num=80, weight=0.6 * u.mS, seed=1)
inh_conn = brainevent.FixedPostNumConn(num_pre=n_inh, num_post=num,
                                       conn_num=80, weight=-6.7 * u.mS, seed=2)

def synaptic_input(spikes):
    spk = brainevent.BinaryArray(spikes)
    g = spk[:n_exc] @ exc_conn + spk[n_exc:] @ inh_conn   # event-driven, unit-aware
    return g

Because the spike vector is a BinaryArray, only neurons that fired this step contribute work — the cost scales with the spike count, not the network size. Swap FixedPostNumConn for a JITCScalarR to scale to networks too large to store explicitly (see Use JIT connectivity for large networks).

Full models (COBA & CUBA)

Complete networks combine this communication layer with neuron and synapse dynamics from the BrainX ecosystem (brainstate, brainpy, braintools). Two reference implementations of the Vogels–Abbott balanced network live in the repository’s examples/ directory:

Example	Synapse model
examples/CUBA_2005.py	Current-based (CUBA) synapses.
examples/COBA_2005.py	Conductance-based (COBA) synapses.

Both use brainevent’s event-driven connectivity under the hood and scale from a few thousand to hundreds of thousands of neurons by changing a single scale factor.

Note

The reference models implement:

• Vogels, T. P. & Abbott, L. F. (2005). Signal propagation and logic gating in networks of integrate-and-fire neurons. J. Neurosci., 25(46), 10786–95.

• Brette, R. et al. (2007). Simulation of networks of spiking neurons: a review of tools and strategies. J. Comput. Neurosci., 23(3), 349–98.