Considering your question and the library in view, my humble comment would be that this library would enable better and faster game programming and physics engines. As a student of ML and data science, I know that ML goes way beyond the ability to vectorize and carry out matrix multiplication. As Andrew Ng put it, there are optimized computation techniques you should not even try to implement yourself unless you are in specialist (maybe Ph.D) in that field. So, while this library will be useful in speeding up some in-browser statistical computations, it will take the arrival of libraries like TensorFlow and Theano on the JavaScript scene to get ML moving there. I am glad this library exists though.