Advanced Spiking Neural Networks – Encoding and Decoding Methods

Cameron Johnson and Ganesh Kumar Venayagamoorthy

Real-Time Power and Intelligent Systems Laboratory

Missouri University of Science and Technology, Rolla, MO 65409, USA

Sponsored by the US NSF EFRI #0836017

Neural Networks (NNs) are well-known function approximators, system identifiers, and pattern recognizers. They are used in many fields and applications when the dynamics of a system change in time such that a static function will not suffice. Problems like identification of power system dynamics require tremendous numbers of inputs and outputs. However, today’s NNs have a significant problem scaling up. The number of neurons required to perform the same task increases faster than the number of inputs, and eventually it simply becomes impossible to have enough neurons with any reasonable computational capacity and expect any level of useable accuracy.

It is known, however, that living neural networks – such as brains – are capable of scaling up seemingly indefinitely. Spiking Neural Networks (SNNs) utilize spiking neuron models which attempt to mimic the behavior of living neurons, and network them in a fashion that allows for the same computational methods used in biological neural networks. It is necessary, however, that real-world information be entered into the SNN, and the results of the SNN’s calculations be extracted. If a method of encoding and decoding real information is developed, a form of SNN known as a polychronous spiking network (PSN) may invert the scaling problem, revolutionizing NN applicability and enabling power system identification of nearly arbitrary-sized systems!

References

  • C. Johnson, G. K. Venayagamoorthy, and P. Mitra, “Comparison of a spiking neural network and an MLP for robust identificaiton of generator dynamics in a multimachine power system,” Neural Networks, Vol. 22, Issues 5-6, July/August 2009, pp. 833 – 841.
  • C. Johnson, G. K. Venayagamoorthy, “Encoding Real Values into Polychronous Spiking Networks,” 2010 IEEE World Congress on Computational Intelligence, July 18-23, 2010, Barcelona, Spain.
  • E. Izhikevich, “Polychronization: Computation with Spikes,” Neural Computation, Vol. 18, pp. 245-282.
  • W. Gerstner, W. Kistler, Spiking Neuron Models. Cambridge University Press, 2002. ISBN 9780521890793