Skip to content

Commit

Permalink
Update note
Browse files Browse the repository at this point in the history
  • Loading branch information
SomeBottle committed Jan 20, 2025
1 parent a7d6f7b commit 061aeb8
Show file tree
Hide file tree
Showing 2 changed files with 6 additions and 0 deletions.
Empty file added .nojekyll
Empty file.
6 changes: 6 additions & 0 deletions ReadingNotes/Research/DiffrentialPrivacyIntroReadingNotes.md
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@
$$
datasets \rightarrow \fbox{Some Process} \rightarrow outputs
$$

中间这个流程可以是:

1. 计算数据的统计数据。
Expand All @@ -40,6 +41,7 @@ Datasets' & & & & \downarrow \\
\fbox{A,B,C,\ ,E} & \rightarrow & \fbox{Magic} & \rightarrow & Outputs'
\end{matrix}
$$

💡 所谓的“**基本一致**”指的是经过相同的流程 $Magic$ ,在两个数据集上的输出**有类似的 ❗ 概率分布**情况。

---
Expand Down Expand Up @@ -105,6 +107,7 @@ $$
\mathbb{P}[A(D_2)=Output] \le e^\varepsilon \cdot \mathbb{P}[A(D_1)=Output]
\end{matrix}
$$

把上面这个式子稍微变一下形就成为了:

$$
Expand Down Expand Up @@ -178,6 +181,7 @@ $\mathbb{P}[A(Yes)=Yes]=0.75$ 是 $\mathbb{P}[A(No)=Yes]=0.25$ 的**三倍**,
* 如果猜测概率是 $0.5$ ,说明攻击者**拿不准主意**

攻击者在拿到经过 $A$ 的输出 $O$ 后,**怀疑目标在 $D$ 中的**的概率会变为:

$$
P\left[D=D_{in}\mid A(D)=O \right]
$$
Expand All @@ -202,6 +206,7 @@ $$
$$
C(D)=(A(D),B(D))
$$

对应输出为一对 $O=(O_A,O_B)$

因此有:
Expand All @@ -213,5 +218,6 @@ P[C(D_1)=O] &= P[A(D_1)=O_A] \cdot P[B(D_1)=O_B] \\
& \le e^{2\varepsilon} P[C(D_2)=O]
\end{align}
$$

组合之后 $C$ 为**满足 $2\varepsilon$ - 差分隐私**的处理机制。

0 comments on commit 061aeb8

Please sign in to comment.