Add eprop work to main tools

[andrewgait]

Incorporate eprop models into the main tools.

This implementation effectively sits alongside the current "standard" neuron models and does not impinge too much on those; the current sticking points (which could be worked on) are:

• It may be necessary to further silo the eprop implementation away from the standard models (for example, creating a spike_processing_eprop.c file and/or a synapses_eprop.c file)
• In a similar vein, the standard Spike Source Poisson C code takes advantage of the multicast packet with payload to deal with cases where multiple spikes may be sent in the same timestep by sending them as one packet with a payload to indicate the number of spikes. This needed to be removed in the current eprop implementation because the communication of the learning signal into the eprop model population requires the use of the multicast packet with payload functionality. We could plausibly have two separate spike source poisson executables to denote these two scenarios.

[coveralls]

coverage: 60.979% (-0.3%) from 61.316% when pulling fa3478a on eprop_adaptive_update into 0f8e5b7 on master.

[Bekkozhan]

Hi all!

I'm working with Andrew Davison at the Institute of Neuroscience Paris-Saclay on extending PyNN to support spike-based learning algorithms such as e-prop and EventProp across neuromorphic backends. I've been reviewing this PR as part of a SpiNNaker feasibility assessment. The implementation looks complete, with two known integration sticking points noted in the description. Is there still active interest in resolving those and getting this merged into the main tools?

Bek

[andrewgait]

Hello @Bekkozhan - this was worked on a number of years ago, and I am not sure how easy it will be to update and merge into the main tools - I don't work on this project any more (though I keep an eye out for things that happen). I'm tagging @rowleya and @Christian-B who are still working on it to see if they have any opinions about whether this branch/PR is worth reviving.

[Bekkozhan]

Thank you @andrewgait - will wait for @rowleya and @Christian-B to comment on this.