Hey craigstuntz, I recently read the paper associated with this library and messed around with it: https://github.com/shaih/HElib. Where does this library sit in with the progress that has been made since with homomorphic encryption (if you’re familiar with it at all), and currently what is the biggest challenge/blocker currently blocking mainstream use of homomorphic encryption if libraries like yours and this are being publicly released like so?
The single biggest barrier to practical use is efficiency. Fully homomorphic encryption in the general case is currently many orders of magnitude slower than other methods. You can mitigate this to some degree by using other methods when homomorphic operations aren’t needed and by optimizing the HE algorithm for the computations to be performed, but even in this case it’s probably still too slow.
There are also concerns about proofs of security in real-world use. Even with a known secure algorithm, it’s harder to implement FHE than non-HE since you have to design the encryption procedure in such a way as to accommodate the computations you intend to perform on the cyphertext. There’s a lot more room for error, since you can’t just take an off-the-shelf algorithm implementation and call “encrypt.”
But efficiency is the show-stopping issue right now.
Author here. Lots has happened since I wrote this. I’m still optimistic about practical, real world use. Any questions, hit me up!
What are your favorite, recent papers about homomorphic encryption?
“Machine Learning Classification over Encrypted Data” [PDF], “Indistinguishability Obfuscation from the Multilinear Subgroup Elimination Assumption” [PDF], and also the (2012) short blog post “Building the Swiss Army Knife”.
Hey craigstuntz, I recently read the paper associated with this library and messed around with it: https://github.com/shaih/HElib. Where does this library sit in with the progress that has been made since with homomorphic encryption (if you’re familiar with it at all), and currently what is the biggest challenge/blocker currently blocking mainstream use of homomorphic encryption if libraries like yours and this are being publicly released like so?
The single biggest barrier to practical use is efficiency. Fully homomorphic encryption in the general case is currently many orders of magnitude slower than other methods. You can mitigate this to some degree by using other methods when homomorphic operations aren’t needed and by optimizing the HE algorithm for the computations to be performed, but even in this case it’s probably still too slow.
HElib requires multiple minutes and GBs of RAM for basic cryptographic operations, according to the authors.
There are also concerns about proofs of security in real-world use. Even with a known secure algorithm, it’s harder to implement FHE than non-HE since you have to design the encryption procedure in such a way as to accommodate the computations you intend to perform on the cyphertext. There’s a lot more room for error, since you can’t just take an off-the-shelf algorithm implementation and call “encrypt.”
But efficiency is the show-stopping issue right now.
thank you for the reply. Very informative!