Skip to content

Commit

Permalink
Merge pull request #34 from datawhalechina/sluds-theorem
Browse files Browse the repository at this point in the history
add proof for slud's inequality
  • Loading branch information
zhimin-z authored May 27, 2024
2 parents 1869098 + 8dc5931 commit 3b17972
Showing 1 changed file with 24 additions and 4 deletions.
28 changes: 24 additions & 4 deletions docs/chapter1/chapter1.md
Original file line number Diff line number Diff line change
Expand Up @@ -690,10 +690,30 @@ $$
$$
P(\frac{X}{m} \geq \frac{1}{2}) \geq \frac{1}{2}\left[1 - \sqrt{1-\exp\left(-\frac{m\varepsilon^{2}}{1-\varepsilon^{2}}\right)}\right]
$$
其中$p = (1- \varepsilon)/2$。

该定理的证明使用了正态分布的标准尾边界,其所需前序知识超出了本笔记的讨论范围,详细证明可参考[论文](https://projecteuclid.org/download/pdf_1/euclid.aop/1176995801)

其中$p = (1-\varepsilon)/2$。

$Proof.$
二项随机变量$X$统计在$m$次独立伯努利试验中成功的次数,成功概率为$p$。对于对于大的$m$,二项分布$B(m,p)$可以近似为均值$\mu=mp$和方差$\sigma^2=mp(1-p)$的正态分布:
$$
\begin{aligned}
\mu &= \frac{m(1-\varepsilon)}{2} \\
\sigma^2 &= \frac{m(1-\varepsilon^2)}{4}
\end{aligned}
$$
令$Z=\frac{X-\mu}{\sigma}$,代入$\mu$和$\sigma$,有:
$$
P[\frac{X}{m} \geq \frac{1}{2}] = P[Z \geq \frac{\frac{m}{2}-\mu}{\sigma}] = P[Z \geq \frac{\varepsilon\sqrt{m}}{\sqrt{1-\varepsilon^2}}]
$$
根据正态分布不等式(定理 20),有:
$$
P[Z \geq x] \geq \frac{1}{2}\left[1 - \sqrt{1-\exp\left(-\frac{2x^2}{\pi}\right)}\right] \geq \frac{1}{2}\left[1 - \sqrt{1-\exp\left(-x^2\right)}\right]
$$
代入可得:
$$
P[Z \geq \frac{\varepsilon\sqrt{m}}{\sqrt{1-\varepsilon^2}}] \geq \frac{1}{2}\left[1 - \sqrt{1-\exp\left(-\frac{m\varepsilon^2}{1-\varepsilon^2}\right)}\right]
$$
得证。



## 定理 18: Johnson-Lindenstrauss 引理
Expand Down

0 comments on commit 3b17972

Please sign in to comment.