A new algorithm for matrix multiplication has been discovered by Google's DeepMind, breaking the 50-year-old record set by Strassen's algorithm. AlphaTensor, a game playing program for matrix decomposition, was used to discover thousands of new fast algorithms, including ones for five by five matrices in modulo-2. While the technology does not make mathematicians obsolete, it provides a valuable tool that helps researchers find new results and guide their intuition. The potential for human and artificial intelligence collaboration is an exciting frontier that is only now being fully explored.
1. Matrix multiplication is a powerful and complex mathematical operation that appears in various fields.
2. It's a fundamental operation in mathematics that can be performed on two-dimensional arrays of numbers called matrices.
3. Researchers have been seeking more efficient ways to multiply matrices together.
4. Volker Strassen discovered a new algorithm in 1969 to multiply two by two matrices that requires only seven multiplication steps, beating the previous record of eight steps.
5. DeepMind, a research lab for artificial intelligence, discovered a new algorithm using reinforcement learning that beat Strassen's algorithm for multiplying two four by four matrices in modulo-2, where the elements are only zero or one, breaking the 50-year-old record.
6. AlphaTensor, a program created by DeepMind, was trained to decompose 3D tensors in different ways to achieve the fewest multiplication steps in matrix multiplication algorithms.
7. AlphaTensor rediscovered Strassen's algorithm within minutes of training.
8. AlphaTensor discovered thousands of new fast algorithms, including ones for five by five matrices in modulo-2.
9. The collaboration between human and artificial intelligence can empower people to do more.