Training spiking neural networks with a controller and STDP

This is an ongoing research project at the Institute of Neuroinformatics (INI) of the university of Zürich and ETH Zürich. This paper showed how to train neurons and networks using a controller coupled with spike-timing-dependent plasticity (STDP). We try to extend this to spiking neural networks, with more realistic performance benchmarks (e.g. MNIST) and deeper networks, to see if this is indeed a viable training algorithm.

The work is in progress. Code is in Python, based on pytorch and pytorch-lightning.

Credits

• Previous code this was based on: Pau Vilimelis Aceituno, Sander de Haan
• New code in pytorch and pytorch lightning: Martino Sorbaro, Sander de Haan
• Extension, debugging, new experiments on deep networks: Alexander Efremov
• Supervision: Pau Vilimelis Aceituno, Benjamin Grewe