Decision boundary that maximizes the margin, defined as the distance between the hyperplane and the nearest data points from each class (support vectors).
Homoscedasticity