Optimal Pruning for Neural Architectures using Fisher Information Distances

A new neural network pruning scheme is introduced using geodesic distance in model space based on the Fisher information metric, outperforming magnitude and…