DeepMind's AlphaTensor AI Discovers Faster Matrix Multiplication Methods
DeepMind introduces AlphaTensor, an AI system that discovers novel, efficient, and provably correct algorithms for matrix multiplication. This breakthrough solves a 50-year-old mathematical puzzle and has significant implications for computing speed across various applications.
DeepMind introduces AlphaTensor, the first artificial intelligence system designed to discover novel, efficient, and provably correct algorithms for matrix multiplication. This breakthrough solves a mathematical puzzle that remains open for over 50 years regarding how to multiply two matrices as quickly as possible. Built upon the foundation of AlphaZero, the agent famous for superhuman performance in board games like chess and Go, AlphaTensor expands this technology from playing games to solving complex mathematical problems.
Matrix multiplication is a fundamental task that impacts the overall pace of a large number of computations in modern technology. Despite being a simple algorithm taught in high school, it plays a critical role in processing images on smartphones, recognizing verbal commands, generating graphics for video games, and running neural networks. Because developing computing hardware that multiplies matrices effectively consumes vast amounts of resources, even small improvements in efficiency yield significant real-world benefits.
The automated discovery of these algorithms is highly complex due to the vast number of possible combinations, but AlphaTensor demonstrates that machine learning has the potential to go beyond human intuition. The research represents a major milestone in DeepMind's objective to advance science and use AI to solve fundamental problems. The team officially publishes this groundbreaking study in the prestigious journal Nature.