Limits to high-speed simulations of spiking neural networks using general-purpose computers.

Friedemann Zenke,Wulfram Gerstner

doi:10.3389/fninf.2014.00076

Abstract

To understand how the central nervous system performs computations using recurrent neuronal circuitry, simulations have become an indispensable tool for theoretical neuroscience. To study neuronal circuits and their ability to self-organize, increasing attention has been directed toward synaptic plasticity. In particular spike-timing-dependent plasticity (STDP) creates specific demands for simulations of spiking neural networks. On the one hand a high temporal resolution is required to capture the millisecond timescale of typical STDP windows. On the other hand network simulations have to evolve over hours up to days, to capture the timescale of long-term plasticity. To do this efficiently, fast simulation speed is the crucial ingredient rather than large neuron numbers. Using different medium-sized network models consisting of several thousands of neurons and off-the-shelf hardware, we compare the simulation speed of the simulators: Brian, NEST and Neuron as well as our own simulator Auryn. Our results show that real-time simulations of different plastic network models are possible in parallel simulations in which numerical precision is not a primary concern. Even so, the speed-up margin of parallelism is limited and boosting simulation speeds beyond one tenth of real-time is difficult. By profiling simulation code we show that the run times of typical plastic network simulations encounter a hard boundary. This limit is partly due to latencies in the inter-process communications and thus cannot be overcome by increased parallelism. Overall, these results show that to study plasticity in medium-sized spiking neural networks, adequate simulation tools are readily available which run efficiently on small clusters. However, to run simulations substantially faster than real-time, special hardware is a prerequisite.

Highlights

Neurons communicate with each other by short electrical pulses, called action potentials or spikes, which can be considered as unitary events
To compare the possible simulation speed of balanced network models using different publicly available simulators and standard hardware (Methods) we adapted the conductance based VogelsAbbott network (Vogels and Abbott, 2005), which has been used as a benchmark in the past (Brette et al, 2007; Sharp and Furber, 2013)
In particular we implemented the same network in Neuron (Carnevale and Hines, 2006), NEST (Gewaltig and Diesmann, 2007), Brian (Goodman and Brette, 2008) and our own simulator Auryn (Methods)

Summary

Introduction

Neurons communicate with each other by short electrical pulses, called action potentials or spikes, which can be considered as unitary events. The dynamics of a single neuron, which forms one unit of a large brain network, are relatively simple. Even if each neuron is described as a relatively simple dynamic processing unit (e.g., an adaptive integrate-and-fire neuron with two or three update equations per neuron Izhikevich, 2003; Brette and Gerstner, 2005; Gerstner et al, 2014), the sheer number of units suggests that faster than real-time simulation of these equations will be hard to achieve on a single core. The evolution of these synaptic variables depends on activity of both the sending and the receiving neuron so that their treatment requires additional care and readily available parallelization approaches cannot be used. The changes in the dynamic values associated with the synaptic contact points are referred to as synaptic plasticity

Methods

Results

Conclusion