1
$\begingroup$

From solving a hard-margin SVM primal problem we get: $$ w = \sum{\alpha_i y_i x_i} \\ \sum{a_i y_i} = 0 $$

Where $\alpha$ is the lagrangian multiplier vector. After solving for $w$ (using the dual problem) we can also solve for $b$ (the bias) with one of the support vectors by $$y_i(wx_i+b)=1$$

Given that, I'm looking for a way to prove that $$\lVert w \rVert ^2 = \sum{\alpha_i}$$

Thanks.

$\endgroup$
1
$\begingroup$

Multiply both sides of the third expression with $\alpha_i$ and sum $\forall i$: $$A=\sum_i \alpha_iy_i(w^Tx_i+b)=\sum_i\alpha_i$$ RHS is what we want, the LHS can be evaluated as: $$A=w^T\left(\sum_i{\alpha_i y_i x_i}\right)+b\sum_i{\alpha_i y_i}=w^Tw+0=||w||^2$$

$\endgroup$

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.