Parameter estimation --- 
We know the density functin for some reason to believe.
What we don't know is the parameter of the density function.

To get a bird's eye view, we assume here 1-D
examples of such density functions with parameter are


Now our data is x= 3, 7, 2, 9.
\subsubsection{maximul-likelyhood}

\subsubsection{Bayesian parameter estimation}

\subsection{Non-paremetric density estimation}

1) density function $p(\mbox{\bf x} \vert \omega)$ is known
2) high dimensional density function can be represented as the product of one-dimensional
function

=> Let's make a more realistic assamption, "We don't know the distribution of samples, i.e.,
we don't know the form of the density function.

Let's estimate $p(\mbox{\bf x} \vert \omega)$ from our sample patterns.

Or, let's estimate $P(\omega \vert \mbox{\bf x})$ directly from our sample patterns.
The typical example is {\it Nearest neighbor rule.}


\subsubsection{Parzen-Window classifier}

\subsubsection{Nearest-Neighbor classifier}