We can fit a linear decision boundary corresponding to the high-dimensional feature space in a Feature Matrix X without explicitly calculating X.įrom a tea ching context, it is historically common in the Machine Learning community to introduce Mercer Kernels paired with Support Vector Machines (SVMs). From the perspective of Machine Learning, Mercer Kernels can be viewed as performing a type of “semi-automated” feature engineering on a set of “basis” variables in a Design Matrix. In fact, such feature spaces can even be infinitely dimensional (as we will show). Photo by James Harrison on Unsplash 1: Introductionįor linear smoothers and linear-predictor based sampling estimators, Mercer Kernels are a highly convenient tool for fitting linear decision boundaries in high dimensional feature spaces.
