Comments on: Millionaire’s Problem

By: Samrobb

Samrobb — Sat, 17 Nov 2007 07:11:40 +0000

I’ve thought about the problem (”Is X interested in Y”) in terms of sparse bloom filters, with an agreement on how data is canonicalized before it is hashed. This allows you to construct a binary blob that you can hand to me, and that I can then query, without knowing exactly what data went in to making up the blob. It still requires some OOB data - but that’s an agreement on protocol, not subject.

By: The Doctor What

The Doctor What — Mon, 05 Nov 2007 21:46:33 +0000

Re: the hashing for album collections:

How about using shingles instead? They can be compared. It’s used in search engines a lot.

Re: millionares problem

I remember a paper that came out around 1992 or so. The problem was to do the classic “flip a coin, call it in the air” with two remote people and prevent lying. It was similar, except that I actually understood (eventually) the math.

Bug:

STEP 2: Bob takes the C value he computed and adds his wealth value J and subtracts one, C - J + 1 = 749 - 6 + 1 = 744. Bob then sends Alice this value.

I assume that should have the words “adds” and “substracts” flipped…

Ciao!

By: Bryan Mills

Bryan Mills — Wed, 31 Oct 2007 20:24:07 +0000

Agreed, the interesting item to know is that I own a Neil Diamond album or just that I have some interest in X. Using the music taste example its really not important that you know that I own an album but more important to know that people (aggregate) that own similar albums as you do also own some other album you don’t have. This means you need a way of grouping people without revealing identities.

I’ve looked at are partial hashing of collections in the hope of grouping multiple people together. You can’t use exact hashing because I might own 1 album different than you which would cause us not to be grouped. If you do partial hashing of your collection of music then you can group with one another and then having a mapping to other hash values that contain other albums. The hash works as just a way of aggregating preferences in an anonymous way. I actually have a working prototype of this and will try to post this soon.

The problem with my little prototype is it current looks at all permutations of your album collection to and creates hash entries for each. I’ve looking at using two hash functions and creating a 2d space where you then map your entries into that instead of the 1d hash function. Then you can do things like range queries over the 2d space that should help cut down the number of permutations needed. This still needs some more thought, although if your interested you should look at how you can do range queries using a DHT.

By: Samrobb

Samrobb — Mon, 29 Oct 2007 17:58:11 +0000

This is near, but it took me a while to identify something that was bugging me about it: it’s simply a way to compare two numbers. The fact that we’re comparing monetary amounts is OOB data that needs to be communicated and agreed on between the two parties. I think that it’s generally the OOB data that’s the interesting part of the interaction. Going back to a previous post of yours.. I don’t really care how many Neil Diamond albums you own. The fact that you do own a Neil Diamond album, on the other hand, is an interesting bit (literally) of data. In order to ask/answer that question, though, we still need to agree on the subject.

For me, an interesting problem is figuring out how to answer those binary questions without revealing additional information, *including* the OOB information about the subject.