MIT researchers have improved homomorphic encryption computations to almost (mass) usable levels. Could this be ported to safe? Even needed in a safe computation environment? The implications of this are pretty staggering.
When the maidsafe team adds compute to the network this is essentially the same thing.
That’s really what I’m getting at. Will safe compute have the same level of security? Will it use something like homomorphic encryption? Will it just be an individual command eg: mov aex so it won’t matter about the command because it’s so small? Is it even fleshed out yet?
In order to run computations on a dataset, you either have to give up the data map for that set (lose all privacy of it), use some type of encryption as listed in the article, or have a LOT of overhead managing it all in the backend.
Being that everything is encrypted and chunked client side it sounds very similar to their proposal already, also incentivized. How the gears will work I don’t know or if they already have a model for this but when I read anything about compute being added as a feature it sounds as though it just needs to be added. There’s prolly low level differences between the two by quite a bit but it sounds like they take the same basic steps to achieve distributed computing. Also why give up the whole data map? Why not just a piece of it? Then your client put it back together after computation.
The difference is the way that they’re encrypted. (they being MIT and safe) safe data is useless without all the parts, whereas the MIT data, each part is usable by itself for the purpose of computation.
In short, the way safe data is is encrypted, if you run computation on one chunk and change it, you destroy the data. In the MIT encryption, you run a computation on a chunk and change it, once you have the rest, it will go back together and give you the result.
Okay I get what you’re saying. Have you looked in irvines github? There is something in there for distributed computing I believe. the ‘bud’ repo https://github.com/dirvine/bud/blob/master/README.md
I’ll look into this. Thanks both for the links.
Wow, this is pretty huge. What those two MIT researchers essentially did is prove the concept of SAFE’s approach to distributed private cloud computing. The data layer of the SAFE network is more than suitable for this, it is bloody perfect! MIT’s approach of storing metadata on the blockchain is in terms of security inferior to storing it on SAFE, which also provides obfuscation.
There is a little difference from what I’m getting. If you have a lot of calculations that need to be done, and you send them to 10 others (in little parts of instructions) they could find out that you’re doing some effects on a video. Because some video-parts might be part of the instructions (like: these are video-frames, add this logo to it). From what I get, the MIT folks have fixed this problem, so before sending out the instructions/data that needs to be calculated, it first get’s encrypted. So now you can pay someone to do calculations, on a set of data that is completely encrypted to them.
Ha, great isn’t it. These projects would fit like magic if it can be done. Even while it’s maybe 10 times slower, you just pay some groups some Safecoin to do it for you. Let them create consensus on the computation as well. That would even be more perfect (although a bit slower).
In the past David said that zkSnark was a more promising avenue, due to poor performance of Homomorphic encryption I think. I wonder if this changes that.
whoa there, you’d be calling the red phone.
MIT has been receiving millions of dollars annually and a large part of the federal funding comes from the military
I’m sure a number of those millions are going directly to Blockchain technology and it does take at least US$300 million @ year to keep the blockchain’s lights on.
I watched a recent conference video from MIT and Standford and the focus seemed very much around law, contracts etc, these guys are dreaming of the control Blockchain can provide…and if you read Martin Armstrong you would know that the world is run by lawyers and all they know, is how to write more law…they have no real world experience or knowledge of how the world economy operates and are currently driving everyone off a cliff.
Hello, i have a question about homomorphic encryption, why addition and
multiplication only make any scheme fully homomorphic while as i know
division is not supported, please can one show me how to obtain or which
scheme can support division and floating point operations ?
Given the way that data are encrypted and chopped into pieces, I was wondering
if a homomorphic encryption scheme could ever be implemented for the SAFE network?
Having the possibility to not only store but also process data in encrypted form would
undoubtedly have great benefit. But would it be possible to implement homomorphic
encryption on top of the network? Or is that fundamentally incompatible with the way
that data is handled in the current design?
At the moment it would be incompatible, but it is a very interesting angle if we considered for instance medical records etc. I have not checked for a while now, but the primitives available were very limited last I did look. Worth keeping and eye on this though, I agree.