Pruning neural networks without any data by iteratively conserving synaptic flow

Open in new window