SpikeNNS represents an extension of SNNS - Stuttgart Neural Network Simulator for the simulation of spiking neural networks. The neural model implemented is based on a simplified version of the Spike Response Model (Gerstner, 1999). Neurons are simulated with a limited number of parameters that include: postsynaptic potential, threshold, noise, delays, refractoriness. The SpikeNNS was designed to produce biologically inspired models of cognitive phenomena based on a spike-coding neural model.
The current version of SpikeNNS can be used for the simulation of spiking self-organizing maps and multi-layer architectures with biologically inspired topologies and spike-timing dependent learning. The simulator is based on an event-driven simulation engine that has been carefully designed for the reduction of computational overload in the presence of high neural activities. However, the simulator was not aimed and was not tested for large scale networks simulation (bigger than 10 k units). As part of SNNS, regarded as a research tool with visualization facilities, user friendly interface and a large number of learning frameworks, SpikeNNS is rather aimed at exploring operating principles of the brain functioning on the base of abstracted models.
What is Stuttgart Neural Network Simulator?
Stuttgart Neural Network Simulator (SNNS) is a software simulator,
currently available for Unix and Windows platforms, developed since
1990 at the Institute for Parallel and Distributed High Performance
Systems (IPVR) at the University of Stuttgart (Zell et al., 1992). It
supports arbitrary network topologies, it is highly configurable and
includes a relatively large number of learning procedures, such as
backpropagation algorithms, ART maps, Kohonen networks, time-delay and
recurrent networks. The graphical user interface offers a 2D/3D
representation of the neural networks and allows a user-friendly
control of the kernel during the simulation run. The sources of C
implementation for Unix platforms are available to download from SNNS web
site and they can be easily extended with user-defined libraries.
SNNS is a widely distributed neural network simulator and its use and
development is technically supported by the SNNS team and by the SNNS
discussion mailing list.
The advantages of using the SNNS framework for the development of a pulsed neural network simulator are twofold. First, there is a certain benefit from using the substantial functionality and visualization capacities of SNNS simulator. Second, SpikeNNS extension can be of interest to the SNNS user community. There were also limits in adapting a simulator built for processing with rate coding neurons for the simulation of spiking neurons behavior. In order to build SpikeNNS we needed to re-engineer parts of the kernel and to add dedicated functions, in order to support computation with spiking neurons. The library of spiking neurons functions implement: the Spike Response Model (SRM), the event-driven simulation engine, functions for the application of time-coded input patterns, routines for learning with spiking neurons.
SpikeNNS is currently implemented as a C library in Linux, part of the SNNSv4.2 kernel for Unix platforms. It has access to SNNS functionality and to the graphical user interface (GUI). Hence, we believe it is suitable for use by both non-programmers and software programmers. SpikeNNS is distributed as a patch file, consisting of a library of C functions that can be added to SNNS original version. SNNS functionality and models are preserved in SpikeNNS
[ intro
| download
| general features
| examples
| documentation
| papers ]
