A Neuromorphic Network for Generic Multivariate Data Classification

This project implements the neuromorphic classifier network as described in [1].

In its current version, it requires the "Spikey" neuromorphic hardware system [2], that is developed at Kirchhoff-Institute for Physics, Heidelberg University [3].

How to run an example:

$ cd src
$ python mnist_classifier_on_spikey.py --help
Usage: mnist_classifier_on_spikey.py [options]

Options:
  -h, --help            show this help message and exit
  -s WORKSTATION, --station=WORKSTATION
	                spikey workstation to use (default: None)
  -n NUM_DATA_SAMPLES, --num_data_samples=NUM_DATA_SAMPLES
	                total number of data samples to use (default: 200)
  -d DIGITS_TXT, --digits=DIGITS_TXT
	                digits to be used (default: 5,7)
  -o OUTPUT_FILE, --output_file=OUTPUT_FILE
	                put detailed results in this file (default: None)
  -r, --retrain_VRs     train a new Neural Gas instead of reusing default VRs
	                (Default: False)
  --save_spiketrains=SAVE_SPIKETRAINS
	                specify a file in which to store the spike trains. The
	                spiketrains will be stored in a file named
	                'spiketrains.pkl' as a pickled dictionary, with keys
	                'train' and 'test', each containing a list of
	                dictionaries that containing the actual spike trains.
	                Default is not to store spike trains.

For example, if you want to clasify digits 4 and 8, using 300 samples per digit, and write the output into a file called "4_8_300.txt", you'd use

$ python mnist_classifier_on_spikey.py -n 300 -d 4,8 -o 4_8_300.txt -r
[...]

The above will only work if you have access to a 'Spikey' chip. If you'd rather like to run it using a simulator, and are willing to help in the porting process, please raise an issue with the repository.

Porting shouldn't be that difficult if you have some PyNN experience. I'll be glad to assist.

References

[1] Schmuker, M.; Pfeil, T.; Nawrot, M. P. A Neuromorphic Network for Generic Multivariate Data Classification. Proc. Natl. Acad. Sci. U. S. A. 2014, 111, 2081–2086. http://www.pnas.org/cgi/doi/10.1073/pnas.1303053111 .

[2] Pfeil, T.; Grübl, A.; Jeltsch, S.; Müller, E.; Müller, P.; Petrovici, M. A.; Schmuker, M.; Brüderle, D.; Schemmel, J.; Meier, K. Six Networks on a Universal Neuromorphic Computing Substrate. Front. Neurosci. 2013, 7, 11. http://dx.doi.org/10.3389/fnins.2013.00011 .

[3] http://www.kip.uni-heidelberg.de/cms/vision/projects/facets/neuromorphic_hardware/single_chip_system/the_spikey_chip/