Original title: Evolutionary algorithms as an alternative to backpropagation for supervised training of Biophysical Neural Networks and Neural ODEs
Authors: James Hazelden, Yuhan Helena Liu, Eli Shlizerman, Eric Shea-Brown
The article explores training neural networks with accurate neuron models to understand brain circuit organization. Investigating the standard learning method, stochastic gradient descent through backpropagation (BP), the researchers find it unstable for models needing stiffness and high nonlinearity.
Original article: https://arxiv.org/abs/2311.10869