二类分类问题 AdaBoost 的训练误差界

最新推荐文章于 2023-04-05 17:11:08 发布

skmygdrs

最新推荐文章于 2023-04-05 17:11:08 发布

阅读量128

点赞数

文章标签：分类机器学习深度学习

本文链接：https://blog.csdn.net/skmygdrs/article/details/124866234

版权

$Z_m\leqslant \exp \left( -2{\gamma _m}^2 \right) \\Z_m=\sum_{i=1}^N{w_{mi}\exp \left( -\alpha _my_iG_m\left( x_i \right) \right)} \\ =\sum_{y_i=G_m\left( x_i \right)}{w_{mi}\exp \left( -\alpha _m \right)}+\sum_{y_i=-G_m\left( x_i \right)}{w_{mi}\exp \left( \alpha _m \right)} \\ =\left( 1-e_m \right) e^{-\alpha _m}+e_me^{\alpha _m} \\ \alpha _m=\frac{1}{2}\log \frac{1-e_m}{e_m} \\ \left( 1-e_m \right) e^{-\alpha _m}+e_me^{\alpha _m}=\left( 1-e_m \right) \left( \frac{e_m}{1-e_m} \right) ^{\frac{1}{2}}+e_m\left( \frac{1-e_m}{e_m} \right) ^{\frac{1}{2}} \\ =2\left( e_m\left( 1-e_m \right) \right) ^{\frac{1}{2}} \\ =\left( 1-4*\left( \frac{1}{2}-e_m \right) ^2 \right) ^{\frac{1}{2}}$

$Z_m=\left( 1-4{\gamma _m}^2 \right) ^{\frac{1}{2}} \\ 0<e_m<1\Rightarrow -\frac{1}{2}<\gamma _m<\frac{1}{2}\Rightarrow 4{\gamma _m}^2<1 \\ \left( 1-x \right) ^{\frac{1}{2}}=1-\frac{1}{2}x-\sum_{n=2}{\frac{2n-3}{2^nn!}x^n} \\ -\frac{1}{2}\left( 1-x \right) ^{-\frac{1}{2}} \\ -\frac{1}{4}\left( 1-x \right) ^{-\frac{3}{2}} \\ -\frac{3}{8}\left( 1-x \right) ^{-\frac{5}{2}} \\ e^{-\frac{x}{2}}=1-\frac{1}{2}x+\sum_{n=2}{\frac{\left( -1 \right) ^n}{2^nn!}x^n} \\ -\frac{1}{2}e^{-\frac{x}{2}} \\ \frac{1}{4}e^{-\frac{x}{2}} \\ -\frac{1}{8}e^{-\frac{x}{2}} \\ \left( 1-x \right) ^{\frac{1}{2}}\,\,<\,\,e^{-\frac{x}{2}}\Rightarrow \left( 1-4{\gamma _m}^2 \right) ^{\frac{1}{2}}<\exp \left( -2{\gamma _m}^2 \right)$