What is Homomorphic Encryption, and Why Should I Care? | Hacker News Reader

What is Homomorphic Encryption, and Why Should I Care?

blogs.teamb.com

34 pointsskorks11 comments

There might be some interesting uses for this in financial markets as well. One of the major problems in financial markets is that if you've got a large position you need to buy/sell you can't just put an order out in the market because you'll move the market price.

One of the more popular ways of solving this at the moment is through the use of "dark pools", third parties who will match up counterparties anonymously who want to do large trades in opposite directions. However this isn't ideal because going to a third party introduces latency and complexity with credit checking, etc.

If you could have some way of allowing firms to place orders with each other directly without exposing the details of the orders, it would have a major impact.

You might work at the institutional level where they want to do this frequently, but I don't think most players would trade directly with each other. One of the purposes of exchanges is to reduce counterparty risk. So only those firms that have the resources to fully vet each other would do this. Not all firms will do this because then you have to vet everyone you trade with which is time consuming.

The many reasons why institutions don't often trade directly with each other, including the anonymity, are covered in the book "Trading and Exchanges: Market Microstructure for Practitioners"

Sure but in most markets I'm guessing the majority of flow is between institutions who already have credit lines with each other (or alternatively use clearing houses like CLS or the DTCC).

There are already equity dark pools which operate on the basis that everyone has credit lines with everyone else rather than taking on the counterparty risk/margin management themselves.

Very cool, I'm definitely "staying tuned" on this, as I don't believe it would require any change to the actual computational software. Immediate, effective encryption for cloud computing, regardless of the platform? Sign me up.

The article he's basing the work on is rather simple, though I haven't given it a hard deconstruction. An encryption / decryption algorithm research article that also includes derivation and proofs, in 25 pages? That's downright easy compared to many I've seen.

If this can be made to work the way these articles are suggesting I think it will be revolutionary. I think about some business that has my personal information on file for the purpose of maintaining an account. It could be the phone company, credit card whatever. I would feel better if all that was encrypted so that not even the programs that operate on that data would be able to use it for ID theft.

Like the author says it will be cool to upload encrypted data to the cloud for computation without worrying about the data being compromised but it will be a while before enough people learn the math to start working on practical implementations.

Practical implementations of the algorithm, or are you thinking that this would require a change in the cloud computing platform (which it shouldn't, which is part of why it's cool)?

If it's the former, that's what the article is about. The blogger is going to be building the algorithm described in the second paper.

This totally blows my mind. I'm sorry I have nothing terribly useful to add, but the idea that you could do meaningful work on data without really ever knowing the data is... well, it's weird. :)

would this help netflix? i'm thinking it might do marginally, but that with a large enough dataset you only need equality to start identifying values (for example, you can take the encrypted text in the "state" field that occurs most often, and assume that is the most populous state, etc etc).

You may be right, I'm not sure. I wonder if it would be possible to take all the data, put it into a text file (or equivalent) and encrypt the whole thing.

The problem I see here is that while you are trying to develop the best netflix prediction strategy, it is often necessary to look at and understand the data itself.

yes, that's the only way it could help (encrypting the whole thing), and then not only do you have the problem of not being able to see the data, you also have an efficiency issue (which, even in some future state where this is improved enough to be usable, is still going to be a real drag).

i was thinking that somehow encrypting each piece of data individually would help, but the kind of processing netflix requires (and the kind that make it possible to extract data based on statistical properties) isn't doing calculations with the values, but doing statistical processing of the tokens (which is why only equality is needed, really). so encryption of individual items doesn't change much.

Reply on news.ycombinator.com