《神经网络与深度学习》习题答案

前言

自己写的或找的习题答案，其中自己写的答案都会同步更新到邱锡鹏老师在 github 上关于这本书的 solutions 项目中（具体题解都在 issues 中）。

第 11 章

11-1

(1)

$$已知p(x_1,x_2,x_3)=p(x_3)\cdot p(x_2|x_3)\cdot p(x_1|x_2)$$

• 独立性
  $$\begin{array}{rl}p(x_1,x_2,x_3)& =p(x_2|x1,x3)\cdot p(x_1,x_3)\\ p(x_3)\cdot p(x_2|x_3)\cdot p(x_1|x_2)& =p(x_2|x1,x3)\cdot p(x_1,x_3)\\ p(x_1,x_3)& =\frac{p(x_3)\cdot p(x_2|x_3)\cdot p(x_1|x_2)}{p(x_2|x1,x3)}\\ & \ne p(x_1)\cdot p(x_3)\\ \therefore X_1与X_3不独立& \end{array}$$

• 条件独立性

$$\begin{array}{rl}p(x_1,x_3|x2)& =\frac{p(x_1,x_2,x_3)}{p(x_2)}\\ & =\frac{p(x_3)\cdot p(x_2|x_3)\cdot p(x_1|x_2)}{p(x_2)}\\ & =\frac{p(x_2,x_3)\cdot p(x_1|x_2)}{p(x_2)}\\ & =\frac{p(x2)\cdot p(x_3|x_2)\cdot p(x_1|x_2)}{p(x_2)}\\ & =p(x_1|x_2)\cdot p(x_3|x_2)\\ \therefore X_1\perp \perp X_3|X_2& \end{array}$$

(2)

与 (1) 镜像，不再赘述。

(3)

$$已知p(x_1,x_2,x_3)=p(x_2)\cdot p(x_1|x_2)\cdot p(x_3|x_2)$$

• 独立性
  $$\begin{array}{rl}p(x_1,x_2,x_3)& =p(x_2|x1,x3)\cdot p(x_1,x_3)\\ p(x_2)\cdot p(x_1|x_2)\cdot p(x_3|x_2)& =p(x_2|x1,x3)\cdot p(x_1,x_3)\\ p(x_1,x_3)& =\frac{p(x_2)\cdot p(x_1|x_2)\cdot p(x_3|x_2)}{p(x_2|x1,x3)}\\ & \ne p(x_1)\cdot p(x_3)\\ \therefore X_1与X_3不独立& \end{array}$$

• 条件独立性

$$\begin{array}{rl}p(x_1,x_3|x2)& =\frac{p(x_1,x_2,x_3)}{p(x_2)}\\ & =\frac{p(x_2)\cdot p(x_1|x_2)\cdot p(x_3|x_2)}{p(x_2)}\\ & =p(x_1|x_2)\cdot p(x_3|x_2)\\ \therefore X_1\perp \perp X_3|X_2& \end{array}$$

(4)

$$已知p(x_1,x_2,x_3)=p(x_1)\cdot p(x_3)\cdot p(x_2|x_1,x_3)$$

• 独立性
  $$\begin{array}{rl}p(x_1,x_2,x_3)& =p(x_2|x1,x3)\cdot p(x_1,x_3)\\ p(x_1)\cdot p(x_3)\cdot p(x_2|x_1,x_3)& =p(x_2|x1,x3)\cdot p(x_1,x_3)\\ p(x_1,x_3)& =\frac{p(x_1)\cdot p(x_3)\cdot p(x_2|x_1,x_3)}{p(x_2|x1,x3)}\\ & =p(x_1)\cdot p(x_3)\\ \therefore X_1与X_3独立& \end{array}$$

• 条件独立性

$$\begin{array}{rl}p(x_1,x_3|x2)& =\frac{p(x_1,x_2,x_3)}{p(x_2)}\\ & =\frac{p(x_1)\cdot p(x_3)\cdot p(x_2|x_1,x_3)}{p(x_2)}\\ & \ne p(x_1|x_2)\cdot p(x_3|x_2)\\ \therefore X_1/\perp \perp X_3|X_2& \end{array}$$