Score:1

Proof of Basic Composition in Differential Privacy

id flag

I'm currently reading the proof of basic composition from the paper https://link.springer.com/content/pdf/10.1007/11761679_29.pdf. In particular, Theorem 1 in Section 2.2.

The proof starts as follows:

enter image description here

My question is why we can assume the set $S$ is in the form $S_1\times S_2\times\ldots\times S_T$. In general, I remember that for differential privacy, we need to prove the inequality for all sets $S$.

I don't believe the statement holds without the assumption, and I don't think we proving it for this case implies the general case (there are counter examples even in the discrete case).

Mark avatar
ng flag
Are they assuming that $S$ takes that form? I would assume they meant implicitly that $S_i$ may depend on both $i$, and $x_1,\dots, x_{i-1}$, i.e. $S_i$ could more precisely be written as $S_{i, x_1, \dots, x_{i-1}}$. I think their proof still goes through in this setting, and makes no assumption that $S = S_1\times \dots \times S_T$.
George Li avatar
id flag
You are completely right. Since they wrote it as a product, it felt like assuming independence and hence being in the form $S_1\times\cdots\times S_T$, but that is indeed not the case. Thanks so much!
mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.