CSCI 3151 - M16 - Regularization: L2, L1, and beyond скачать в хорошем качестве

CSCI 3151 - M16 - Regularization: L2, L1, and beyond

This module introduces regularization as a principled way to control model capacity and improve generalization in supervised learning. Using concrete code examples in Python, we compare unregularized models with L2 (ridge), L1 (lasso), and elastic net penalties, showing how regularization affects training vs validation error, coefficient magnitudes, and sparsity. We connect these ideas back to the bias–variance tradeoff, discuss how to tune the regularization strength in practice, and briefly touch on probabilistic and deep-learning perspectives (e.g., weight decay). By the end, students should understand when and why to use different regularizers, and how these choices impact both performance and interpretability. Course module page: https://web.cs.dal.ca/~rudzicz/Teaching/CS...