A nconcept in Supervised learning

Intuition of functional margin

Definition of functional margin of hyperplane (where $y^{(i)} \in \{-1, 1\}$, remember) wrt to data point;

def of functional margin wrt to entire training set, is just the minimum of the functional margin over all data points. We need a normalization.

The formula is

$\gamma^{(i)} = y^{(i)}[(\frac{\omega}{||\omega||})^T x^{(i)} + \frac{b}{||\omega||}]$

It is very similar to Geometric margin