I don’t believe this will work? I would have to see an actually working example though. With actual data, not matrix vector multiplications those are trivial.
Doing math on garbled numbers and then reverse garbling it? Easy. Doing text parsing on garbled text? Probably impossible, but I’d loveto be proven wrong. I also think you have to reveal what kind of functions you want used?
The homomorphism in category theory is often shown by a commutative diagram, where you can go from a point to another by interchanging the order of operations. In the below diagram for FHE, you can go from (a, b) to E(a*b) in two separate ways.
It works in the sense that the operations are performed on binary numbers, so text handling works the same way it normally does assuming the handler function is encrypted to match. Once you have multiplication and addition, you can make logic gates and general computing follows from there - although with the noise being amplified thru each logic gate, the more complex the functions the more bootstrapping is required and the less I see this being doable in the short term.
For a working example, check out apple’s homomorphic encryption page, they use it for landmark identification and afaik will be using it for siri whenever they get to that update. It’s slow but it’s already usable - I’m not personally convinced it’ll be used everywhere, but the technology is super cool and I hope it shows up more
assuming the handler function is encrypted to match.
Yeah, this is the thing I’m doubting / don’t understand how that would work.
E.g. A* / navigation problems.
You send private start and goal points.
Either the stuff is truly private, then the program can’t read it.
Or the program can read it, but then the owner of the machine the program runs on can just read it from memory.
It doesn’t matter if it says “45124x5234234fgasdgf” or “Paris”, because the program state will identify that. Even if you encrypt the entire location database (with stuff that’s then fully known to the server) and it will still look up “45124x5234234fgasdgf” and the server can trivially decrypt that.
check out apple’s homomorphic encryption page
Interesting, but I’m more leaning on “they have a vested interest to lie about this” rather than “surely this is correctly working tech that keeps me safe”. Like Amazons “AI supermarket” that was just a bunch of indians doing video surveillance.
And their explanation makes the same amount of sense as the blog post. I have no doubt that it can work for simple commutative math operations, over “smooth” domains. Where my doubt comes in is functions where the encryption would cause the operation to take place outside of the domain bounds.
How does an encrypted asin or acos work?
Anyway, thanks for the answer, I was recently impressed by GNU Taler, which also did something cryptographic stuff I didn’t think was possible. So I’m not saying this is heresy and can’t be done and trying to say it will work is forbidden, I just don’t think the explanations so far are detailed enough.
I don’t believe this will work? I would have to see an actually working example though. With actual data, not matrix vector multiplications those are trivial.
Doing math on garbled numbers and then reverse garbling it? Easy. Doing text parsing on garbled text? Probably impossible, but I’d loveto be proven wrong. I also think you have to reveal what kind of functions you want used?
Even in math this doesn’t work for all problems.
It works in the sense that the operations are performed on binary numbers, so text handling works the same way it normally does assuming the handler function is encrypted to match. Once you have multiplication and addition, you can make logic gates and general computing follows from there - although with the noise being amplified thru each logic gate, the more complex the functions the more bootstrapping is required and the less I see this being doable in the short term.
For a working example, check out apple’s homomorphic encryption page, they use it for landmark identification and afaik will be using it for siri whenever they get to that update. It’s slow but it’s already usable - I’m not personally convinced it’ll be used everywhere, but the technology is super cool and I hope it shows up more
Yeah, this is the thing I’m doubting / don’t understand how that would work.
E.g. A* / navigation problems.
You send private start and goal points.
Either the stuff is truly private, then the program can’t read it.
Or the program can read it, but then the owner of the machine the program runs on can just read it from memory.
It doesn’t matter if it says “45124x5234234fgasdgf” or “Paris”, because the program state will identify that. Even if you encrypt the entire location database (with stuff that’s then fully known to the server) and it will still look up “45124x5234234fgasdgf” and the server can trivially decrypt that.
Interesting, but I’m more leaning on “they have a vested interest to lie about this” rather than “surely this is correctly working tech that keeps me safe”. Like Amazons “AI supermarket” that was just a bunch of indians doing video surveillance.
And their explanation makes the same amount of sense as the blog post. I have no doubt that it can work for simple commutative math operations, over “smooth” domains. Where my doubt comes in is functions where the encryption would cause the operation to take place outside of the domain bounds.
How does an encrypted asin or acos work?
Anyway, thanks for the answer, I was recently impressed by GNU Taler, which also did something cryptographic stuff I didn’t think was possible. So I’m not saying this is heresy and can’t be done and trying to say it will work is forbidden, I just don’t think the explanations so far are detailed enough.