algorithm - How does SVM work? -
is possible provide high-level, specific explanation of how svm algorithms work? high-level mean not need dig specifics of different types of svm, parameters, none of that. specific mean answer explains algebra, versus solely geometric interpretation. understand find decision boundary separates data points training set 2 pre-labeled categories. understand seek finding widest possible gap between categories , drawing separation boundary through it. know how makes determination. not looking code, rather explanation of calculations performed , logic. know has orthogonality, specific steps "fuzzy" everywhere find explanation.
here's video covers 1 seminal algorithm quite nicely. big revelations me (1) optimize square of critical metric, giving value that's positive, minimizing square (still differentiable) gives optimum; (2) using simple, not-quite-obvious "kernel trick" make vector classifications compute easily.
watch @ how unwanted terms disappear, leaving n+1 vectors define gap space in n dimensions.