Pruning neural networks, which involves removing a fraction of their weights, can often
maintain high accuracy while significantly reducing model complexity, at least up to a certain …