Optimization and machine learning
The bridge between the two disciplines: a learning model is, in the end, an optimization problem to be solved with the same solver as the rest of the course. A QP that classifies and a QP that estimates.
-
Support Vector Machine
The SVM as a convex QP: maximum margin, dual formulation, support vectors and kernels — with no machine learning library.
-
Least-squares regression
Least squares as a convex QP: the KKT conditions are the normal equations, the comparison with robust and quantile regression, ridge.