Recall Guassian models

• k classes: each class gets its own weight Pi_j and guassian P_j

  $$ log(\pi_j) + log(P_j(x) \text{ quadratic} $$

Each class has its own quadratic function

When a new point x arrives:

• each class evaluates its function on x
• predict largest value’s label

Multiclass Linear Classification

Multiclass Logistic Regression

Multiclass Perceptron

Multiclass SVM