GFDL

Abstract

This is a Python library that provides infrastructure for users to train gradient-free (neural network) machine learning models in a variety of problem domains. The avoidance of gradients/backpropagation is achieved by the randomization and fixing of a subset of neural network weights and biases during training.