ReSuMe, Chronotron and SPAN all use STDP-like local learning rules to implement their training algorithm (though they approach the training differently, e.g. SPAN uses gradient descent via spikes transformed into analogue signals through kernel convolution whereas ReSuMe does not to my understanding). The papers I've read claim that they are not suitable for training multi-layer spiking neural networks and can only be used to train single-layer networks[0][1], but it is not entirely clear to me why and getting a straight answer has proven difficult. Is it because they use local learning rules and propagating weight changes across multiple layers is impossible?

[0] Ponulak, Filip & Kasiński, Andrzej. (2011). Introduction to spiking neural networks: Information processing, learning and applications. Acta neurobiologiae experimentalis. 71. 409-33.

[1] Kasabov, N. K. (2018). Time-Space, Spiking Neural Networks and Brain-Inspired Artificial Intelligence (Springer Series on Bio- and Neurosystems). 1st. Springer Publishing Company, Incorporated. ISBN: 3662577135.

  • $\begingroup$ This is not a question about Theoretical Computer Science, and therefore off-topic here. $\endgroup$ Commented Dec 6, 2021 at 8:45
  • $\begingroup$ I disagree - the methods are particular implementations but the question applies to any theoretical approach that uses STDP. $\endgroup$ Commented Dec 6, 2021 at 21:39


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.