Cryptography is a very exact (albeit heuristic, at the core, but nonetheless exceedingly exact) science. Asking if he has "seen the memory required for that" a bit insulting and telling. A level of detail far, far deeper than that is absolutely required for any useful analysis of a paper like that.
I really do like the idea, @Tom, so much that I've already analysed it and the fact of the matter is that a Bloom filter is necessarily a factor of $\frac{1}{\log(2)}$ larger than the raw list of hashes for the same security target. Any smaller bloom filter is trivially insecure due to a higher false positive rate.
@Tom It's such a trivial result that I'm having trouble finding a paper dedicated to it. I can give you a whole stack of papers that all consider it "obvious"
ok so I'll assume the total number of bloom taps is given by $k=\frac{m}{n}\log(2)$ because it minimizes the false positive rate
Basically it's one of the trivial details that nobody should have to explain, you can convince yourself using extremely basic algebra techniques to minimize the false positive rate.
It's well known in industry that bloom filters are a factor of $\frac{1}{\log(2)}$ away from optimal
I'm wondering, do you know the number of hash functions and bits used in the bloom filter will be chosen to meet the desired false probability rate for only a fraction of the 'public hashes' mentioned in the paper? for example: 1024 items per bloom filter, and so k=14 and m=19631
I looked at it. It did look friendlier with the psuedo code, though I will admit I have not seriously spent the time getting into whatever you're working on.
When it comes to those disposable signatures, to me it seems like the problem is barely about cryptography anymore is and is more about managing state/revocation
(which is a good thing, to have the hard part taken care of)
I just think it's a novel method that can** be easily understood and can perform well, meaning it can be used in practice because many people can develop/use it, which isn't the same for the more complex schemes