Efficient generation of connectivity in neuronal networks from simulator-independent descriptions

Djurfeldt, Mikael; Eppler, Jochen M.; Davison, Andrew P.

doi:10.3389/fninf.2014.00043

Journal Article

FZJ-2014-04540

Efficient generation of connectivity in neuronal networks from simulator-independent descriptions

Djurfeldt, M. (Corresponding Author) ; Davison, A. P. ; Eppler, J. M.FZJ*

2014
Frontiers Research Foundation Lausanne

Frontiers in neuroinformatics 8, 43 (2014) [10.3389/fninf.2014.00043] special issue: "Python in Neuroscience II"

This record in other databases:

Please use a persistent id in citations: http://hdl.handle.net/2128/7904 doi:10.3389/fninf.2014.00043

Abstract: Simulator-independent descriptions of connectivity in neuronal networks promise greater ease of model sharing, improved reproducibility of simulation results, and reduced programming effort for computational neuroscientists. However, until now, enabling the use of such descriptions in a given simulator in a computationally efficient way has entailed considerable work for simulator developers, which must be repeated for each new connectivity-generating library that is developed. We have developed a generic connection generator interface that provides a standard way to connect a connectivity-generating library to a simulator, such that one library can easily be replaced by another, according to the modeler's needs. We have used the connection generator interface to connect C++ and Python implementations of the previously described connection-set algebra to the NEST simulator. We also demonstrate how the simulator-independent modeling framework PyNN can transparently take advantage of this, passing a connection description through to the simulator layer for rapid processing in C++ where a simulator supports the connection generator interface and falling-back to slower iteration in Python otherwise. A set of benchmarks demonstrates the good performance of the interface.

Classification: