A novel AI-acceleration paper presents a method to optimize sparse matrix multiplication for machine learning models ... which are often performed on vector processors or custom matrix engines.