Comparing Rewinding and Fine-tuning in Neural Network Pruning