VIVA LA RESISTANCE
[Interesting]
October 16, 2022
Multiplication of two 4x4 matrices usually consists of 64 steps. Volker Strassen in 1969 came up with an algorithm that reduces the number of steps to 49. The AlphaTensor neural network from DeepMind found a way to reduce the number of steps to 47. And in 5x5 matrix multiplication, the neural network reduced the number of steps from 98 to 95. Considering that this is one of the most popular operations in modern computers, video cards and other devices will soon become a little faster.