SynFlow: Pruning neural networks without any data by iteratively conserving synaptic flow

The Lottery Ticket Hypothesis has shown that it's theoretically possible to prune a neural network at the beginning of training and still achieve g...