8. Loss Functions for Regression and Classification
We start by discussing absolute loss and Huber loss, two alternative to the square loss for the regression setting, which are more robust to outliers. Next, we introduce our approach to the classification setting, introducing the notions of score, margin, and margin-based loss functions. We discuss basic properties of the hinge loss (i.e SVM loss), logistic loss, and even the square loss, considered as a margin-based loss. The interplay between the loss function we use for training and the properties of the prediction function we end up with is a theme we will return to several times during the course.
Access the full course at https://bloom.bg/2ui2T4q