genn-team / ml_genn

A library for deep learning with Spiking Neural Networks (SNN).
https://ml-genn.readthedocs.io
GNU Lesser General Public License v2.1
24 stars 7 forks source link

Time constant issues with e-prop output learning rule #100

Open neworderofjamie opened 4 months ago

neworderofjamie commented 4 months ago

The e-prop output learning rule should use a presynaptic trace filtered with tau_out (kappa = e^(-dt/tau_out):

image

But, in our implementation it uses a presynaptic trace filtered with tau_mem. Fix should simply be to calculate kappa at https://github.com/genn-team/ml_genn/blob/master/ml_genn/ml_genn/compilers/eprop_compiler.py#L498 and pass this through to the learning rule.