DeepMind Matrix Multiplication and Moore’s Law
A few days ago, Google announced it had accomplish a 50-year breakthrough in the matrix multiplication through the use of its AlphaGo deep learning computer. I took linear algebra in first year engineering and matrix multiplication was not fun. If you wanted to multiple two 4 x 4 matrices, you had to do 64 calculations.
Volker Strassen in the 1960s developed a way to do the same math with only 49 calculations. Now, DeepMind found a way to do it in 47 calculations. Of course, this would seem just a simple little decrease in effort, akin to a Simpsons scene about a cure for 15 stab wounds when they need one for 17:
However, if put into effect, it’s a huge difference potentially shaving lots of effort off of computing. Here, we have an AI tool developing a new tool for developing new tools.
It’s not often we get a direct example of Moore’s Law at work but this is a clear illustration of technology unlocking new technologies.
