Grouping similar weight values together and replacing them with shared cluster centers to reduce model size.