next up previous
Next: The learning problem Up: Support Vector Machines (SVM) Previous: Support Vector Machines (SVM)

Linear separation of a feature space

A hyper plane in an n-D feature space can be represented by the following equation:

\begin{displaymath}
f({\bf x})={\bf x}^T {\bf w}+b=\sum_{i=1}^n x_i w_i+b=0
\end{displaymath}

Dividing by $\vert\vert{\bf w}\vert\vert=\sqrt{{\bf w}^T{\bf w}}$, we get

\begin{displaymath}
\frac{{\bf x}^T {\bf w}}{\vert\vert{\bf w}\vert\vert}=P_{\bf...
...vert\vert{\bf w}\vert\vert}=-\frac{b}{\sqrt{{\bf w}^T{\bf w}}}
\end{displaymath}

indicating that the projection of any point ${\bf x}$ on the plane onto the vector ${\bf w}$ is always $-b/\vert\vert{\bf w}\vert\vert$, i.e., ${\bf w}$ is the normal direction of the plane, and $\vert b\vert/\vert\vert{\bf w}\vert\vert $ is the distance from the origin to the plane. Note that the equation of the hyper plane is not unique. $c f({\bf x})=0$ represents the same plane for any $c$.

The n-D space is partitioned into two regions by the plane. Specifically, we define a mapping function $y=sign(f({\bf x})) \in \{1,-1\}$,

\begin{displaymath}
f({\bf x})={\bf x}^T {\bf w}+b=\left\{ \begin{array}{ll} >0,...
... y=sign(f({\bf x}))=-1,\;{\bf x}\in N \\
\end{array} \right.
\end{displaymath}

Any point ${\bf x}\in P$ on the positive side of the plane is mapped to 1, while any point ${\bf x}\in N$ on the negative side is mapped to -1. A point ${\bf x}$ of unknown class will be classified to P if $f({\bf x})>0$, or N if $f({\bf x})<0$.

Example:

svm1.png

A straight line in 2D space ${\bf x}=[x_1, x_2]^T$ described by the following equation:

\begin{displaymath}
f({\bf x})={\bf x}^T {\bf w}+b=[x_1,x_2]
\left[ \begin{array...
...t[ \begin{array}{c} 1  2 \end{array} \right]-1
=x_1+2x_2-1=0 \end{displaymath}

devides the 2D plane into two halves. The distance between the origin and the line is

\begin{displaymath}
\frac{\vert b\vert}{\vert\vert{\bf w}\vert\vert}=\frac{1}{\sqrt{w_1^2+w_2^2}}=\frac{1}{\sqrt{5}}=0.447
\end{displaymath}

Consider three points:


next up previous
Next: The learning problem Up: Support Vector Machines (SVM) Previous: Support Vector Machines (SVM)
Ruye Wang 2016-08-24