Neural networks with many zero or near-zero weights, making them simpler and more interpretable than dense networks.