Efficient parameter calibration and real-time simulation of large scale spiking neural networks with GeNN and NEST

Abstract

Spiking neural networks (SNN) represent the state-of-the-art approach to the biologically realistic modeling of nervous system function. The systematic calibration for multiple free model parameters to achieve robust network function demands high computing power and large memory resources. Special requirements arise from closed-loop model simulation in virtual environments, and from real-time simulation in robotic application. Here, we compare two complementary approaches to efficient large scale and real-time SNN simulation. The widely used NEural Simulation Technology (NEST) parallelizes simulation across multiple CPU cores. The GPU-enhanced Neural Network (GeNN) simulator uses the highlyparallel GPU-based architecture to gain simulation speed. We quantify fixed and variable simulation costs on different hardware configurations. As benchmark model we us a spiking cortical attractor network with a topology of densely connected excitatory and inhibitory neuron clusters. We show that simulation time scales linear with the simulated biological time and, for large networks, approximately linear with the model size as dominated by the number of synaptic connections. Additional fixed costs with GeNN for model compilation are independent of model size while fixed costs with NEST for model building increase linear with model size. We demonstrate how GeNN can be used for simulating networks with up to 3, 5 million neurons (> 3 × 1012 synapses) on a high-end GPU, and up to 250 thousand neurons (25 billion synapses) on a low-cost GPU. Real-time simulation was achieved for networks of more than 90 k neurons. We conclude that GeNN is in advantage for large networks and real-time applications while NEST plays out its strengths in a high degree of flexibility ideal for prototyping, and easy access for non-expert programmers.