Knowledge Maxxing← Back to topic library
ML MaxxingIntermediate75 sec
Regularisation
Regularisation adds a penalty to a model's training objective that discourages overly complex patterns, helping reduce overfitting.
Prerequisite: Bias-variance trade-off
What matters
By penalizing complexity directly during training, regularisation lets a model use more capacity without fully exploiting it to memorize noise in the training data.
Example
A regularised linear model might be pushed to keep its coefficients small unless the data strongly justifies a large one, avoiding overreliance on any single feature.
Common misconception
Regularisation does not eliminate overfitting entirely on its own; it reduces the tendency, but data quality and model choice still matter.
Recall question
What does regularisation add to a model's training objective, and why?
Keywords and related concepts
regularisationoverfittingpenalty termbias-variance trade-off