Microsoft boffins build better crypto for secure medical data crunching As genome research - and the genomes themselves - get passed around the scientific community, the world's woken up to the security and privacy risks this can involve. A Microsoft research quintet has therefore published ways to help scientists work on genomic data while reducing the risk of data theft. The team published an …
First reaction : bollocks. Computers are stupid, in order to modify a given data, they have to read it first.
The I check out the paper itself. Ouch. Way too much math in there, not enough air. Looks solid though, and specifically states that it makes it possible to work on encrypted data without decrypting it.
Now it may well be that I did not find the chink in the process, and it is certain that I didn't understand everything (though I do find amusing that the actual encyption library is written in C++ with a wrapper in C#), but still, that is one serious mathematical paper.
Count my mind boggled.
Same here. Still trying to work my way through the math (it's been a while...), I guess this will be my weekend-read-at-leisure. Very interesting. Must admit though that I first read it as homeopathic encryption and went right into WTF? mode.
One thing though, El Reg: a quntet is a group of five. The paper lists six authors. Admit it, you didn't want to write 'sextette'. (Yeah, I know, 'Tips and corrections'. But that is e-mail, and right now I'm on a machine that doesn't want do that.)
It's... weird.
http://eprint.iacr.org/2014/106.pdf
( The set of plaintext being used ) = ( Real numbers in range [X] )/( (Cyclotomic polynomial X* ) , (an integer) )
* "the unique irreducible polynomial with integer coefficients, which is a divisor of x^n-1 and is not a divisor of x^k-1 for any k < n." Meaning, a constant and unique function that doesn't fit into any lower functions.
Examples (wikipedia):
1: x - 1
2: x + 1
3: x^2 + x + 1
4: x^2 + 1
5: x^4 + x^3 + x^2 + x^1 + x + 1
This is just for the plaintext, mind. The actual encryption process involves taking the set of (decomposition by w, itself * w^n) and does... magic. This magic produces noise, in the form of the equation becoming vastly over-complicated, but it's still an equation and can be multiplied and added just like in grade school algebra.
Additionally, any operations you can perform will always produce some level of noise. There are, however, expensive operations to convert them back to the unreducible form you would get if you started with it, which lets you avoid going over the limit set at the start and wiping some of your data.
The idea itself is 30+ years old (apparently) by now, but it's only recently been made more usable by the fact it doesn't take half an hour to perform a single operation. The exact details of that process is what's being talked about in the paper, and honestly, I have no clue. And I've just spent three hours looking at this.
...Four hours.
The idea itself is 30+ years old (apparently) by now, but it's only recently been made more usable by the fact it doesn't take half an hour to perform a single operation.
Fully homomorphic encryption (FHE) is only six years old, and HE wasn't practical before then, regardless of performance considerations. Prior to the invention of FHE, no one had figured out an HE scheme that permitted both addition and multiplication operations, so its applicability was extremely limited. When you can do both addition and multiplication (in GF(2)), you can perform arbitrary computations - because in GF(2) addition is XOR and multiplication is AND, so you can implement any Boolean circuit.
As for how it works, it's actually not conceptually all that difficult. You encode plaintext chunks as polynomials in one ring. Encryption maps them to polynomials in a second ring. Addition and multiplication operations on that second set of polynomials changes them, but as long as you don't do too much of it, you can be sure that when you do the reverse mapping (decryption) you get back to the polynomials that represent those same operations performed on the plaintext.
It's a bit like error-correcting codes based on groups: as long as the error ("noise") doesn't get too large, you stay in the neighborhood of where you want to be.
Typically, for generality, FHE schemes employ "bootstrapping": they do some operations, decrypt, and re-encrypt to clear out the noise. This paper shows how to tweak the settings beforehand, if you know what you might want to do to the encrypted data, so you can avoid that.
At any rate, that's my understanding. I looked at this stuff a bit more closely a few years back, and I've just skimmed this paper.
Encryption, like all Win10 services, are covered under Paragraph 14 of Win10's EULA, which houses aka.ms/msa which demands that MSFT slurp all your private offline data, and imposes an arbitrary CODE OF CONDUCT to justify that slurping. So, when someone accuses 'you' of whatever, then the slurping, then all the THIRD PARTY DATA on your machine is slurped, so THEY CAN SUE YOU for data breach.
So I told my physician clients to avoid Win10 and all the other 'services' covered under aka.ms/msa. Unless you like bankruptcy.
So who cares about encryption, when MSFT claims CONTRACTURAL RIGHT to slurp it? And you are left the victim, and it goes scot free?
These guys are Microsoft Research (http://research.microsoft.com/en-us/ ) and are distinct from the main Microsoft.
A lot of the software they release as free software (This gives MS lawyers cardiac arrests).
If you go to a talk by them you will see v.little corporate stuff and a lot of geek-researcher enthusiasm
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
