A reinforcement learning approach based on AlphaZero is used to discover efficient and provably correct algorithms for matrix multiplication, finding faster algorithms for a variety of matrix sizes.
Discovering faster matrix multiplication algorithms with reinforcement learning
Posted in information science