Skip to main content
to regularize

to regularize

7.5
To make a model simpler so it learns general patterns, not noise
  • verb
  • /ˈrɛɡjʊləˌraɪz/
  • Specialized
translation icon : regularizar
  • In machine learning, it is common to regularize models to enhance their performance on unseen data.

Examples

  • In effect, the repeated weighting procedure (10) concentrates the solution in the minimal number of "active" elements of g d, still reproducing the measurements sufficiently well. An additional advantage of using this procedure is the possibility to easily incorporate constraints or regularizing penalties.

    Academic text (2016)
  • The research team decided to regularize their model to prevent overfitting during training.

  • But this is a a way to regularize the problem.

  • We then define the following complexity-regularized optimization problem.

    Academic text (2017)
  • Full and regularized partial correlation matrices were also calculated from these time series.

    Academic text (2018)
  • This reduced establishing correspondence to a well-understood image registration problem, which was solved with a regularized form of optical flow.

    Academic text (2018)
  • The image yielding the least V is accepted as the best (final) image and regularizing parameter at a given iteration.

    Academic text (2016)
  • Timecourses were used to estimate full and regularized partial correlation network matrices using FSLnets.

    Academic text (2018)
  • Specifically, for the case of the linear support vector machine (where L is the so-called hinge loss), the regularized optimization problem is equivalent to the following robust optimization problem.

    Academic text (2017)
  • To improve accuracy, the algorithm was designed to regularize its parameters using L2 penalties.

Synonyms

to penalize
vsto regularize
  • Specialized
3 6.5

Add a cost to a model to discourage it becoming too complex

applies broader methods to limit model complexity beyond adding terms to loss

Antonyms

to overfit
  • Specialized
8.4

Make a model memorize the training data so it does not work well on new data

How Regularized

to regularize
  • Specialized
7.5
to penalize
  • Specialized
3 6.5
to constrain
64 5.6

Surface Forms

regularize infinitive
regularized past participle
regularizing present participle
regularizes third person singular

Morphology

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

Formally derived from 'regular' + '-ize', but the statistical/ML sense ('apply regularization penalties') is a technical extension that is not obvious to learners from the base meaning alone.

Etymology

The word regularize uses the same root regula ('rule') as regular and rule, and in statistics or machine learning it means to add limits or small penalties so a model stays simple and does not 'memorize' random details, so regularizing helps the model behave more reliably by following a simpler pattern.