This is v0.1 of the SSBM benchmark suite, a collection of benchmarks for supervised learning in spiking neural networks. It is meant to make supervised learning techniques in spiking nets more comparable to each other.
When using a benchmark from this suite, please cite the benchmark name, address of this repository (https://github.com/fzenke/ssbm) and version number/commit.
Please see this early version of the suite as an invitation to suggest and add your own benchmarks. Please submit your suggestions through the ticketing system or contact me by email such that I can make you a contributor of this repo.
All precise timing benchmarks contain a spike raster file with the input spikes (input.ras) and the target spike times (target.ras). These files are human readable two-column formatted. The first column contains the firing time in seconds and the second column denotes the id of the unit emitting a spike.
- poisson: 100 Poisson input spike trains with 0.5s duration, single target output spike train with 5 spikes
- poisson10: 100 Poisson input spike trains with 10s duration, and different Poisson target spike trains at different mean firing rates.
- secgen: Sequence generation benchmark. 100 Poisson input spike trains of 1s duration, 100 output units firing in sequence
- auryn: 100 Poisson input spike trains with 3.5s duration, 100 output spike trains resembling the Auryn logo
- monalisa: 1000 Poisson input spike trains with 33s duration, 500 output spike trains resembling Leonardo da Vinci's Mona Lisa
The IBM DVS event-based gesture dataset. http://www.research.ibm.com/dvsgesture/