Training Delays in Spiking Neural Networks
Spiking Neural Networks (SNNs) are a promising computational paradigm, both to understand biological information processing and for low-power, embedded chips. Although SNNs are known to encode information in the precise timing of spikes, conventional artificial learning algorithms do not take this into account directly. In this work, we implement the spike timing by training the synaptic delays in a single layer SNN. We use two different approaches: a classical gradient descent and a direct algebraic method that is based on a complex-valued encoding of the spikes. Both algorithms are equally able to correctly solve simple detection tasks. Our work provides new optimization methods for the data analysis of highly time-dependent data and training methods for neuromorphic chips.
- 1.Bishop, C.M.: Pattern Recognition and Machine Learing. Springer, New York (2006)Google Scholar
- 4.Dayan, P., Abbott, L.F.: Theoretical Neuroscience: Computational and Mathematical Modeling of Neural Systems. The MIT Press (2001)Google Scholar
- 5.Diehl, P.U., Zarrella, G., Cassidy, A., Pedroni, B.U., Neftci, E.: Conversion of artificial recurrent neural networks to spiking neural networks for low-power neuromorphic hardware. In: 2016 IEEE International Conference on Rebooting Computing (ICRC), IEEE, October 2016. https://doi.org/10.1109/icrc.2016.7738691
- 8.O’Connor, P., Welling, M.: Deep spiking networks. arxiv (2016)Google Scholar