Show me one useful software patent that (a) is not "obvious to one skilled in the art", and (b) benefits society by being granted a monopoly. Just one!

Software rarely requires expensive research that would be worth protecting. Rather than enabling a fair market, this takes fairness out of the market.

Software patents are like getting a patent on "Murder story with final revelation of who did it." Maybe add one or two features, like a "detective with hat", etc. In one fell swoop you would be able to own most murder mysteries.

Software (like books, stories, art, etc) is better handled by Copyright law. May the one who actually has a better product win!

Sorry for the rant.

https://patents.justia.com/patent/20230401438#history

> A method, a neural network, and a computer program product are provided that provide training of neural networks with continued fractions architectures.

It's a bit like the propelling device https://patents.google.com/patent/US1331952A/en -- it doesn't invent or patent shoes, or invent or patent springs, but it patents attaching springs to shoes.

It would seem using continued fractions with elliptic curves what the author wants to do, wouldn't be covered.

However, I still think it can still be challenged if someone can show that continued fractions have been used in with NN before. Or even better, maybe pytorch or other open source projects can explicitly reject crap that's patented. If you put your shit in a junk patent, take it out of the project and enjoy yourself, don't spread it around. So, if the authors of the patent are the ones pushing for the inclusion, then someone should challenge that and have it removed unless the patent is withdrawn.

RTFC: Read the Fucking Claims

The claims are the only part of the patent that really matter because those are the only enforceable language. Plus, this is a patent application so the claims have not even been examined yet.

Without commenting on the merits of this patent itself, remember that new applications of existing techniques are still novel, and hence patentable. In fact, if you think about it, almost all inventions and innovations are just applying novel combinations of well-known techniques to new use-cases. (Anything that doesn't fit that definition and introduces genuinely new methods is usually in the "groundbreaking" category.)

So I'd guess "applying old technique to new problem" is probably the case here. I'm no subject matter expert, and there may be prior art that invalidates this, or it may not meet the non-obviousness bar... but when the "200 year old math" came about neural networks were not really a thing.

Ofcourse, patents are trash.

If the objective is to maximize investment by protecting successful results, I don't think our system is doing a very good job.

Also, this might be tangentially related to Fractan. Any Mathematician there? if there's some relation, the patent might be void/null.

https://en.wikipedia.org/wiki/FRACTRAN