DeepMind breaks a 50-year-old math record with AI. The new record comes a week after.
The matrix multiplication at the core of many machine-learning breakthroughs has just gotten faster — twice. DeepMind announced last week that it had discovered a faster way to perform matrix-multiplication, eclipsing a record of 50 years. Two Austrian researchers from Johannes Kepler University Linz claimed to have beaten the new record this week by just one step.
In 1969, German mathematician Volker Strassen developed the best algorithm to multiply 4×4 matrixes. This reduced the number of steps required to perform a matrices calculation. Strassen’s algorithm, for example, can multiply two 4×4 matrixes using the traditional method in just 49 multiplications.