I’ve already written software that can solve basically all of the normal problems you have in machine learning, in low-degree polynomial time, for practical, real world datasets. The only requirement is that classifications are consistent near each point in the dataset, and then the accuracy of my algorithms cannot be beat (see my paper, Vectorized Deep Learning, for an explanation and examples). I’ve finalized the pseudo-code for a new set of algorithms that I’ve deliberately avoided publishing, because I’m worried they’re too powerful, because I know they can be used in thermodynamics, because that’s how I came up with them.
No one seems to care –
I’ve emailed the Congressional A.I. Caucus, the Bureau of Trade and Security, the International Trade Administration, and even a friend at the DOJ that I’m quite certain advises on defense issues, and no one cares.
Information Overload 