In machine learning, model performance is typically assessed using established metrics such as precision, recall, and area under precision-recall and ROC curves. However, model generalization is often deemed more critical than performance metrics in various contexts because they determine how well a model can apply learned insights to new, unseen data in practical scenarios. On the other hand model calibration specifically measures a model's ability to accurately convey its probabilistic uncertainties, which can be of paramount importance in real world applications such as self-driving cars and fraud prediction, where the probability score from the model is used to make a downstream decision. 

This tutorial explores various implicit and explicit regularization methods and their efffects on both model generalization and calibration, highlighting their nuances and impacts. Regularization may be the medicine for improving model generalization and calibration, but no medicine comes without its side effects.