What is SVM (Support Vector Machine)?
SVM, told as a story about a reviewer drawing the line that splits two tenant groups as widely as possible.
The reviewer wants to tell which side each new tenant will land on, ahead of time.Tenants fall into two groups: the quiet side and the noisy side.
A careless line misjudges new tenants that land close to it.The line can't hug either group too closely.
SVM = the reviewer who draws the line farthest from both groups.In SVM, the points closest to the line — the support vectors — set its position.
Drawing the line with "the widest" gap is the whole point of SVM.A wider margin tends to classify new data near the line more reliably.
SVM isn't actually a cloud-infrastructure story.SVM is a classic machine-learning classification algorithm — covered here as an exam-prep topic, but the building metaphor stops here.
SVM is a classification algorithm that draws a boundary with the widest possible margin between two groups. The line is set using only the closest points — the support vectors.
SVM (Support Vector Machine) is a supervised classification technique that finds the hyperplane (decision boundary) separating two classes with the maximum margin. Data that can't be split linearly is handled with the kernel trick, computed as if mapped into a higher-dimensional space. Since this is a classic ML algorithm rather than a cloud-infrastructure topic, this page's building metaphor is intentionally looser than the others.
When grown-ups say it
- Support Vector Machine
- The reviewer splitting two groups the widest. A supervised-learning classification algorithm.
- Decision boundary (hyperplane)
- The line separating the two groups. In higher dimensions, this line becomes a hyperplane.
- Margin
- The gap between the line and the closest points. Wider margins tend to predict new data more reliably.
- Support vector
- The closest points, the ones that actually set the line. Move just these points, and the boundary moves too.
- Kernel trick
- The trick of reshaping the space when a straight line won't do. Lets the boundary curve when it needs to.
- Supervised learning
- Learning from data that already has labeled answers. Like already knowing which tenants are quiet vs noisy, ahead of time.
- Classification
- Sorting into one of a set of groups. SVM basically sorts into two groups.
- Overfitting
- A boundary fit too tightly to the training data. More likely when the margin is drawn too tight.