🛂 Granting legal status 📏 Making uniform 🤖 Simplifying models

regularization

7.3

A method used in machine learning to keep a model simple so it predicts new data better

noun
/ˌrɛɡjʊləraɪˈzeɪʃən/

Specialized

: regularización

To limit overfitting and encourage a sparse, interpretable parameter weighting, we use L1-regularization.
regularization method

David Cecilia uses regularization for that purposes.

Examples

In summary, regularization is an important technique for helping machine-learning algorithms to generalize well.

Academic text (2017)
An important idea for achieving robustness in machine learning is regularization.

Academic text (2017)
And there are two penalty terms here with different regularization parameters.
An internal length was used as a regularization parameter.

Academic text (2018)
To balance the two energy terms, it acts as a regularization parameter with a positive value.

Academic text (2017)
We can view the regularization penalty as a force that "pulls the solution back" from the unpenalized optimum.

Academic text (2017)
The data-fidelity term models the statistics of measurements here, and the regularization term reflects a priori information.

Academic text (2016)
The regularization method added constraints to the model, improving its ability to generalize beyond the training set.
By applying regularization during training, we can ensure that the model performs well on unseen data.
In summary, regularization is an important technique for helping machine-learning algorithms to generalize well.

Synonyms

regularizer

vsregularization

Specialized

8.6

A method that stops a computer model from fitting training data too closely

is the broader technique that focuses on simplifying models and adding constraints

Antonyms

overfitting

Specialized

8.1

A model that learns the training data too well and fails on new data

Surface Forms

regularization singular

regularizations plural

Morphology

regularization = regularize (semi-transparent) = regular + ize + ation

Morphologically derived from 'regular' via 'regularize', but the machine-learning sense ('adding penalties or constraints') is a specialized technical extension that is not obvious to general learners.

Etymology

Regularization comes from regular and the Latin regula meaning 'rule'. In machine learning it means adding rules or limits to a model so it stays simple and does not learn random details, and that is why the word describes methods that keep a model 'regular' and reliable.