Brette et al. 2007 Benchmarks

This directory contains Python implementations of the benchmarks from the FACETS simulator review (Brette et al. 2007) [2]. These benchmarks are based on the Vogels & Abbott network model [1] and are designed to test different aspects of neural network simulation:

• Benchmark 1 (COBA): Conductance-based synapses with integrate-and-fire neurons

• Benchmark 2 (CUBA): Current-based synapses with integrate-and-fire neurons

• Benchmark 3 (HH-COBA): Conductance-based synapses with Hodgkin-Huxley neurons

• Benchmark 4 (CUBA-PS): Current-based synapses with precise spiking

• Benchmark 5 (CUBA-STDP): Current-based synapses with STDP plasticity

All benchmarks create sparsely coupled networks of excitatory and inhibitory neurons which exhibit self-sustained activity after an initial stimulus. Connections within and across both populations are created at random. Both neuron populations receive Poissonian background input.

Files

• brette_et_al_2007_benchmark.py: Shared framework module containing common network building and simulation functions

• coba.py: Benchmark 1 - Conductance-based synapses (COBA)

• cuba.py: Benchmark 2 - Current-based synapses (CUBA)

• hh_coba.py: Benchmark 3 - Hodgkin-Huxley neurons (HH-COBA)

• cuba_ps.py: Benchmark 4 - Precise spiking (CUBA-PS)

• cuba_stdp.py: Benchmark 5 - STDP plasticity (CUBA-STDP)

Usage

Each benchmark can be run directly as a Python script. For example:

python coba.py

Or from the examples directory:

python -m brette_et_al_2007.coba

The benchmarks will print a summary including: - Number of neurons and synapses - Average firing rates for excitatory and inhibitory populations - Building and simulation times

Benchmark Details

Benchmark 1 (COBA)

• Neuron model: iaf_cond_exp

• Synapse model: conductance-based (exponential)

• Spike times: grid-constrained

• Simulation time: 1000 ms

Benchmark 2 (CUBA)

• Neuron model: iaf_psc_exp

• Synapse model: current-based (exponential)

• Spike times: grid-constrained

• Simulation time: 10000 ms (10x longer due to lower computational load)

Benchmark 3 (HH-COBA)

• Neuron model: hh_cond_exp_traub

• Synapse model: conductance-based (exponential)

• Spike times: grid-constrained

• Simulation time: 1000 ms

Benchmark 4 (CUBA-PS)

• Neuron model: iaf_psc_delta

• Synapse model: current-based (delta/voltage jump)

• Spike times: off-grid (precise spiking)

• Simulation time: 10000 ms

Benchmark 5 (CUBA-STDP)

• Neuron model: iaf_psc_exp

• Synapse model: STDP (E->E), static current (others)

• Spike times: grid-constrained

• Simulation time: 2000 ms

• Features: Random initial weights and membrane potentials, STDP plasticity

References

[1]

Vogels TP, Abbott LF. 2005. Signal propagation and logic gating in networks of integrate-and-fire neurons. Journal of Neuroscience. 25(46):10786-10795. https://doi.org/10.1523/JNEUROSCI.3508-05.2005

[2]

Brette R, Rudolph M, Carnevale T, Hines M, Beeman D, Bower JM, et al. 2007. Simulation of networks of spiking neurons: a review of tools and strategies. Journal of Computational Neuroscience. 23(3):349-398. https://doi.org/10.1007/s10827-007-0038-6

Benchmark 1 of the simulator review (COBA)

Benchmark 1 of the simulator review (COBA)

Benchmark 4 of the simulator review (CUBA-PS)

Benchmark 4 of the simulator review (CUBA-PS)

Benchmark 2 of the simulator review (CUBA)

Benchmark 2 of the simulator review (CUBA)

Benchmark 3 of the simulator review (HH-COBA)

Benchmark 3 of the simulator review (HH-COBA)

Benchmark 5 of the simulator review (CUBA-STDP)

Benchmark 5 of the simulator review (CUBA-STDP)

Brette et al. 2007 Benchmark Framework

Brette et al. 2007 Benchmark Framework