About the research I got one little thing: spiking neurons are a part of Neural Engineering Framework. It’s a stupid name. It defines some linear formulas. And the training is done by least square optimization which leads me to linear (ordinary) least squares in linear regression. It’s not like the gradient descent that chips away little pieces. This one reduces the error in one go (this to be clarified) and hyper vectors allow that because similarity and clustering solved on the encoding step.

https://www.jneurosci.org/content/26/14/3667#xref-ref-17-1 Introduction to hyperdimension https://redwood.berkeley.edu/wp-content/uploads/2021/08/Neubert2019_Article_AnIntroductionToHyperdimension.pdf