AlphaTensor - Discover Faster Matrix Multiplication Algorithms with RL - IPAM at UCLA

Explore a groundbreaking lecture on the application of reinforcement learning to discover more efficient matrix multiplication algorithms. Delve into the innovative AlphaTensor system, an AI agent based on AlphaZero, designed to surpass human intuition in algorithm design. Learn how this approach led to the discovery of algorithms outperforming state-of-the-art complexity for various matrix sizes, including an improvement on Strassen's two-level algorithm for 4x4 matrices—a feat unaccomplished in 50 years. Understand the formulation of this mathematical challenge as a single-player game, the key components enabling reinforcement learning to tackle complex mathematical problems, and the versatility of the AlphaTensor framework. Gain insights into the potential widespread impact of enhancing fundamental computational tasks on overall computing speed.

Matej Balog - AlphaTensor: Discover faster matrix multiplication algorithms with RL - IPAM at UCLA