0
$\begingroup$

In Hard margin support vector machine we have equation of Hyper-plane

$W^T*X+b = 0 $, I want to know why for a point $X_n$ not on Hyper-plane

$W^T*X+b = 0$ we take

$|W^T*X_n+b| = 1$

why do we normalize to 1

$\endgroup$
2
$\begingroup$

If I understand your question correctly, you wonder why the boundary hyperplanes of the separating region, defined as $\{x: w^Tx+b=\pm k\}$, have a fixed value for $k$ (namely 1 in most texts).

Short answer: If $k$ is treated as a parameter to be estimated, then the problem becomes underdetermined.

Exhibit: If $k$ would be treated as an unknown parameter, any candidate solution $(w,b,k)$ would imply a set of candidate solutions $\{(\lambda w, \lambda b, \lambda k) : \lambda\in\mathbb R\}$ which all amount to the same practical solution (that is, the same margin region).

Now, given that the loss function for the hard-margin problem is just $f(w)=\|w\|^2\geq 0$, letting $\lambda\downarrow 0$ shows that any separating region would be optimal.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.