AI System Breaks 50-Year-Old Matrix Multiplication Speed Record
DeepMind's AlphaTensor uses reinforcement learning to discover faster matrix multiplication algorithms, surpassing human intuition. The AI breaks a 50-year-old speed record previously held by Strassen's algorithm.
Researchers at DeepMind introduce AlphaTensor, a deep reinforcement learning system that discovers efficient and provably correct algorithms for matrix multiplication. The AI treats this mathematical challenge as a single-player game, where the objective is finding tensor decompositions within a finite factor space. By exploring this enormous space of possible algorithms, the system bypasses traditional human intuition to uncover entirely new computational methods.
The new AI system achieves a historic milestone by improving upon Strassen’s two-level algorithm for 4 × 4 matrices in a finite field. This marks the first time in 50 years that anyone or anything advances beyond the complexity of Strassen's foundational mathematical discovery. AlphaTensor also finds state-of-the-art algorithms for various other matrix sizes, proving its capability to tackle complex computational problems across the board.
Beyond pure mathematical complexity, AlphaTensor demonstrates remarkable flexibility by adapting its discoveries for real-world applications. The system optimizes matrix multiplication for actual runtime on specific hardware and excels at structured matrix multiplication. These results highlight how artificial intelligence accelerates algorithmic discovery and customizes solutions for different practical criteria, impacting everything from neural networks to scientific computing.