LoginSignup
0
0

More than 3 years have passed since last update.

PRML 演習問題2.58 解答

Last updated at Posted at 2020-06-05

問題

(2.226)は指数型分布族では$\mathbf{lng(\eta)}$の負の勾配が$\mathbf{u(x)}$の期待値を示している.(2.195)の2階微分を取ることで

\begin{align*}
\mathrm{-{\nabla}lng(\eta)=E[u(x)u(x)^T]-E[u(x)]E[u(x)^T]=cov[u(x)]}
\tag{2.300}\\
\mathrm{g(\eta){\int}h(x)exp\bigl\{\eta^Tu(x)\bigr\}dx=1}
\tag{2.195}
\end{align*}

を示せ.

方針

(2.195)の2階微分を行うが、一階微分の値が(2.226)で示されているので(2.226)を微分して上記を示す.

\begin{align*}
\mathrm{-\frac{1}{g(\eta)}{\nabla}g(\eta)=g(\eta){\int}h(x)exp\bigl\{\eta^Tu(x)\bigr\}u(x)dx=E[u(x)]}
\tag{2.225}\\
\mathrm{-{\nabla}lng({\eta})=E[u(x)]}
\tag{2.226}
\end{align*}

解答

\begin{align*}
(2.225)と(2.226)から\\
\mathrm{-{\nabla}lng({\eta})=g(\eta){\int}h(x)exp\bigl\{\eta^Tu(x)\bigr\}u(x)dx}\\
\end{align*}

よって

\begin{align*}
\mathrm{-{\nabla\nabla}lng(\eta))={\nabla}g(\eta){\int}h(x)exp\bigl\{{\eta}^Tu(x)\bigr\}u(x)^Tdx+g(\eta){\int}h(x)exp\bigl\{\eta^Tu(x)\bigr\}u(x)u(x)^Tdx}
\end{align*}

右辺の第1項を式変形すると

\begin{align*}
\mathrm{{\nabla}g(\eta){\int}h(x)exp\bigl\{{\eta}^Tu(x)\bigr\}u(x)^Tdx}&=
\mathrm{{\nabla}g(\eta)*g(\eta){\int}h(x)exp\bigl\{{\eta}^Tu(x)\bigr\}u(x)^Tdx}\\
&=\mathrm{{\nabla}lng(\eta)*E[u(x)^T]}\\
&=\mathrm{-E[u(x)]*E[u(x)^T]}
\end{align*}

したがって

\begin{align*}
\mathrm{-{\nabla\nabla}lng(\eta)=-E[u(x)]*E[u(x)^T]+E[u(x)u(x)^T]=cov[u(x)]}
\end{align*}

よって題意は示された

0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
0
0