It was interesting to read how you did a spiking eural network in PyTorch, but it seems your neurons' states are coupled continuously in time, whereas in the brain, it would be the opposite, ie. the spike timing carries information and not the state values.

> backprop engenders STDP

This is backwards I think, but definitely an interesting association to make

There's a cool package called Nengo from Chris Eliasmith's group at UWaterloo which hasn't received as much attention as I think it should: https://www.nengo.ai/nengo-dl/examples/tensorflow-models.htm...

You may be interested to watch this talk:

http://www.cs.nyu.edu/~yann/deep07/videos/mov010.avi

Slides: http://www.cs.toronto.edu/~hinton/backpropincortex2007.pdf