BNBP

Training networks of Hodgkin-Huxley (HH) neurons is hard! STDP is biologically realistic, but it alone does not generally let us teach the networks to do anything particularly useful. With James Hazelden and Daniel Forger, we used stochastic gradient descent through backpropagation to train feedforward networks of HH neurons. We presented this at the NeurIPS 2020 Beyond Backprop workshop, and the code is on github.