# 1.伯努利随机变量的特例

$\begin{array}{}\text{(237)}& P\left(H\left(n\right)\le k\right)=\sum _{i=0}^{k}{C}_{n}^{i}{p}^{i}\left(1-p{\right)}^{n-i}\phantom{\rule{0ex}{0ex}}=\sum _{i=0}^{k}\frac{n!}{i!\left(n-i\right)!}{p}^{i}\left(1-p{\right)}^{n-i}\end{array}$

$\begin{array}{}\text{(238)}& P\left(H\left(n\right)\le \left(p-\epsilon \right)n\right)\le {e}^{-2{\epsilon }^{2}n}\end{array}$

$\begin{array}{}\text{(239)}& P\left(H\left(n\right)\ge \left(p+\epsilon \right)n\right)\le {e}^{-2{\epsilon }^{2}n}\end{array}$

$\begin{array}{}\text{(240)}& P\left(\left(p-\epsilon \right)n\le H\left(n\right)\le \left(p+\epsilon \right)n\right)\ge 1-2{e}^{-2{\epsilon }^{2}n}\end{array}$

$\begin{array}{}\text{(241)}& P\left(|H\left(n\right)-pn|\le \sqrt{n\mathrm{ln}n}\right)\ge 1-2{e}^{-2\mathrm{ln}n}=1-\frac{2}{{n}^{2}}\end{array}$

# 2.伯努利随机变量的一般情况

$\begin{array}{}\text{(898)}& P\left(|\overline{X}-E\left(\overline{X}\right)|\ge t\right)\le 2{e}^{-\frac{2{n}^{2}{t}^{2}}{\sum _{i=1}^{n}\left({b}_{i}-{a}_{i}{\right)}^{2}}}\end{array}$

${S}_{n}={X}_{1}+{X}_{2}+,...,+{X}_{n}$$S_n = X_1+X_2+,...,+X_n$
$\begin{array}{}\text{(899)}& P\left(|{S}_{n}-E\left({S}_{n}\right)|\ge t\right)\le 2{e}^{-\frac{2{t}^{2}}{\sum _{i=1}^{n}\left({b}_{i}-{a}_{i}{\right)}^{2}}}\end{array}$

$E\left({e}^{sX}\right)\le {e}^{\frac{{s}^{2}\left(b-a{\right)}^{2}}{8}}$

$P\left({S}_{n}-E\left({S}_{n}\right)\ge t\right)=P\left({e}^{s\left({S}_{n}-E\left({S}_{n}\right)\right)}\ge {e}^{st}\right)\phantom{\rule{0ex}{0ex}}\le {e}^{-st}E\left({e}^{s\left({S}_{n}-E\left({S}_{n}\right)\right)}\right)\phantom{\rule{0ex}{0ex}}={e}^{-st}\prod _{i=1}^{n}E\left({e}^{s\left({X}_{i}-E\left({X}_{i}\right)\right)}\right)\phantom{\rule{0ex}{0ex}}\le {e}^{-st}\prod _{i=1}^{n}E\left({e}^{\frac{{s}^{2}\left({b}_{i}-{a}_{i}{\right)}^{2}}{8}}\right)\phantom{\rule{0ex}{0ex}}=exp\left(-st+0.125{s}^{2}\sum _{i=1}^{n}\left({b}_{i}-{a}_{i}{\right)}^{2}\right)$

$g\left(s\right)=-st+0.125{s}^{2}\sum _{i=1}^{n}\left({b}_{i}-{a}_{i}{\right)}^{2}$$g(s)= -st+0.125s^2\sum_{i=1}^n(b_i-a_i)^2$，则$g\left(s\right)$$g(s)$为二次函数，当$s=\frac{4t}{\sum _{i=1}^{n}\left({b}_{i}-{a}_{i}{\right)}^{2}}$$s = \frac{4t}{\sum_{i=1}^n(b_i-a_i)^2}$时函数获得最小值。因此：
$P\left({S}_{n}-E\left({S}_{n}\right)\ge t\right)\le {e}^{-\frac{2{t}^{2}}{\sum _{i=1}^{n}\left({b}_{i}-{a}_{i}{\right)}^{2}}}$

# 3.集成学习的错误率上界

$\begin{array}{}\text{(1168)}& P\left(H\left(n\right)\le k\right)=\sum _{i=0}^{k}{C}_{n}^{i}\left(1-ϵ{\right)}^{i}{ϵ}^{n-i}\end{array}$

$P\left(集成分类错误\right)=P\left(H\left(n\right)\le \frac{n}{2}\right)\phantom{\rule{0ex}{0ex}}=\sum _{i=0}^{\frac{n}{2}}{C}_{n}^{i}\left(1-ϵ{\right)}^{i}{ϵ}^{n-i}\phantom{\rule{0ex}{0ex}}\le exp\left(-\frac{n}{2}\left(1-2{ϵ}^{2}\right)\right)\phantom{\rule{1em}{0ex}}\left(由\epsilon =\frac{1}{2}-ϵ可得\right)$

• 广告
• 抄袭
• 版权
• 政治
• 色情
• 无意义
• 其他

120