ANNarchy: a code generation approach to neural simulations on parallel hardware

Abstract

Many modern neural simulators focus on the simulation of networks of spiking neurons on parallel hardware. Another important framework in computational neuroscience, rate-coded neural networks, is mostly difficult or impossible to implement using these simulators. We present here the ANNarchy (Artificial Neural Networks architect) neural simulator, which allows to easily define and simulate rate-coded and spiking networks, as well as combinations of both. The interface in Python has been designed to be close to the PyNN interface, while the definition of neuron and synapse models can be specified using an equation-oriented mathematical description similar to the Brian neural simulator. This information is used to generate C++ code that will efficiently perform the simulation on the chosen parallel hardware (multi-core system or graphical processing unit). Several numerical methods are available to transform ordinary differential equations into an efficient C++code. We compare the parallel performance of the simulator to existing solutions.

Spike: A GPU Optimised Spiking Neural Network Simulator

“…In order to execute SNN simulations, a range of simulators have emerged with each offering a unique combination of speed, hardware integration, and ease of model definition. In this study we compared the ANNarchy (Vitay et al, 2015), Auryn (Zenke and Gerstner, 2014), Brian2 (Stimberg et al, 2014), GeNN (Yavuz et al, 2016), and NEST (Linssen et al, 2018) simulators against the Spike simulator. These comparisons showed the efficacies of GPU based SNN simulators of which GeNN and Spike are examples.…”

“…These kinds of models may be used to simulate the 'spiking' dynamics of real neurons in the brain, which communicate with each other by emitting electrical pulses called action potentials or 'spikes'. In this pursuit, a number of SNN simulators have emerged (Vitay et al, 2015;Zenke and Gerstner, 2014;Stimberg et al, 2014;Yavuz et al, 2016;Linssen et al, 2018), which provide combinations of high speed execution, hardware support, and simplicity when defining models.…”

PyRates – A Python Framework for rate-based neural Simulations

“…Different software tools have been developed for neural modeling that offer various solutions to render numerical simulations more efficient (e.g. TVB [11], DCM [12], Nengo [13], NEST [14], ANNarchy [15], Brian [16], NEURON [17]). Since the brain is an inherently highly parallelized information processing system (i.e.…”

“…its central and graphical processing units (CPUs and GPUs). Neural simulation tools that implement such mechanisms include ANNarchy [15], Brian [16], NEURON [18], Nengo [13] and PCSIM [19], for example. Each of these tools has been build for neural models of a certain family.…”

“…This means 730 that the projection operation will be performed regardless of whether a spike is 731 produced or not, leading to considerable increases in computation time for large, densely 732 connected single cell networks. Hence, when dealing with neuroscientific questions that 733 implicate the use of spiking neuron models, we currently recommend to use simulation 734 tools such as Nengo [13], NEST [14], ANNarchy [15], Brian [16] or NEURON [17]. Such 735 questions may involve problems where specific spike-timings have a non-negligible 736 influence, where dendritic tree architectures are important or, generally, where the 737 variable of interest looses its meaning when averaged over time or over many neurons.…”