It's "private" "cloud" computing all over again, this time brought to you by jensen et al.
Google is making private AI practical with homomorphic encryption
291–300 of 305 posts
Re: Google is making private AI practical with homomorphic encryption
#292Re: Google is making private AI practical with homomorphic encryption
#293Encryption or not, if it's on somebody else's server, it isn't yours. I don't believe Google has my best interest.
With Fully Homomorphic Encryption it's nobody elses. The basic idea of of the project is to remove the need for trust.
It was not my intent to undercut cryptography as a field or in its use. Consider client side scanning for photos before they're encrypted and uploaded to the cloud. I see no difference AI's future. Content, maybe encrypted, but a form of client-side scanning, or the model itself, will report users.
Re: Google is making private AI practical with homomorphic encryption
#294Earlier quoted context omitted.
the basic encryption scheme used here is fairly straightforward actually, at least the symmetric encryption version. Let s be a uniformly random, 512-dimensional u32 vector. To encrypt a message m (say a 512-dimensional bit vector for simplicity), you 1. generate a 512 x 512 random (u32) matrix A, and 2. generate a 512-dimensional rounded (to the nearest integer) Gaussian, say of standard deviation 10, e. The ciphert…
Will the new (summed) A, e and b be the same size as the originals, and is m2 + m2 still a 512-dimensional bit vector? I though (when I tried to understand it) that some part of the HE inflates some component of the result?
only temporarily, and only for multiplication. At a very high level, the idea is that you view C := [A, b] as satisfying
CS = 2^8 m + e
here, S = [-s, 1] is a padded version of the initial secret. So recast everything as a linear equation (matrix) equation
CS = 2^8m + e
Without getting into too much details, one can define a "product" * such that
(CC)(SS) = (2^8m + e)(2^8m + e)
This becomes a "degree 2" equation. Mildly faking the details for simplicity, one can expand it out not in terms of A, b, but in terms of three components A, b, c, where c is the "degree 2" component. So here things have inflated. But there is also a technique to shrink this back down to a linear equation.
This shrinking process requires some auxiliary data, namely an encryption of SS under S. it is not the problematic part of HE though. Instead, data movement (say a circular rotation by k indices) also requires some "fixing up", though here involving an encryption of rot^i(S) under S.
This is more problematic, as there are many different rotations (often on the order of thousands), and you naively need a piece of auxiliary data for each of them (vs one for multiplication). There are ways to shrink the required number of keys, but in general they're the "heavyweight" part of FHE.
Re: Google is making private AI practical with homomorphic encryption
#295I really want nobody to waste cycles on my AI than me. This is useless because now I need to pay for wasted inference AND encryption cycles on top of wire encryption.
While encrypted AI chatbots seem within reach, we don’t see them as the primary target today. There is a set of untapped applications, such as inter-bank transfers, industrial use cases, and healthcare; where regulations, privacy laws, and compliance requirements prevent institutions from directly accessing or processing sensitive data. This is where we believe Belfort can make a real difference: enabling organizatio…
I think that encrypted/homomorphic inference has overstated benefits and won't work as well for financial/ regulatory cases, where explainability is a bigger problem and AI has that accuracy issue.
Re: Google is making private AI practical with homomorphic encryption
#296This is the same Google that doesn't have e2ee on their password manager by default. Like WTF, it's a password manager.
what would e2ee for a password manager even be? This reads like "Whoa, this encryption key isn't encrypted, we should encrypt it for added security"
Same as any E2EE product. You don't have to trust the server and only the users hold the keys.
Re: Google is making private AI practical with homomorphic encryption
#297> Google is making private
(in the voice of George Bush Jr.) And it feels good doing it.
Google doing something private is like Microsoft doing something secure.
Re: Google is making private AI practical with homomorphic encryption
#298Earlier quoted context omitted.
Yeah. Google is an ad business. Their entire motive for getting invested in AI is ad revenue. We're supposed to believe they just... won't turn on the money fountain? After going into the red for their data center investments? Hell nah. I'd expect this to be something like the Google Ad ID: technically separated from what Google considers personal information, but trivially easy to tie back to an individual person an…
This blog post is about an open source project that you can look at for yourself and decide whether it's suitable for your use case.
Sure. But it will be trickled down into project by google which will effect us all and we won't have a say.
Re: Google is making private AI practical with homomorphic encryption
#299My master's thesis is on a topic in this field (Privacy Preserving ML) and from my understanding HE and other techniques have very high overheads(~10^3) on inference tasks and thus aren't very commercially viable.
The article conspicuously fails to go into much detail about that. I poked around with an AI a bit (to rapidly cover all the linked pages) and it seems the best numbers we can get are from this arxiv paper: https://arxiv.org/html/2506.18150v4 Which says: "We evaluate HE-LRM on UCI (health prediction) and Criteo (click prediction), achieving inference latencies of 24 seconds on UCI and 228 to 489 seconds, respectively…
With GPU acceleration, that 228 to 489s becomes a fraction of a second :)
https://belfortlabs.com/blog/belfort-partners-with-lg-on-enc...
Re: Google is making private AI practical with homomorphic encryption
#300Earlier quoted context omitted.
and you don't know who I am. I might have been closer to that work than you would know.
Why don't you say what they are wrong about then?
every single word. this is a project to enable private AI use cases which is a very real concern for a lot of enterprise customers dealing with highly sensitive data (think about healthcare and police work for example) and has nothing to do in any way shape or form with ads. The guy probably used to work or still works at Google but doesn't mean he understands what he's talking about.