What is regularization, why do we use it, and what are some common methods?
Whether you connect it to the bias-variance trade-off rather than listing techniques.
Machine learningStandard
What a strong answer does
Regularization is how we stop a high-capacity model from treating the training set as the truth rather than as a noisy sample from the truth. If the model is flexible enough, it can fit accidental quirks: mislabeled examples, sampling artifacts, rare correlations, or just random noise. That often improves training accuracy while making test performance worse. Regularization deliberately gives up some freedom, and often some training-set performance, in exchange for a model whose learned pattern is more likely to survive on unseen data.