TERMIUM Plus^®

Fiche 1,  Justifications,  Anglais

Record number: 1, Textual support number: 1 DEF

An a priori assumption or information that a learning algorithm uses to predict the output of an input it has never encountered.  3, fiche 1, Anglais, - inductive%20bias

Record number: 1, Textual support number: 1 CONT

Informally, an inductive bias can be defined as the preference of a model for one hypothesis that is consistent with the training data over another. Inductive biases encourage the model towards this hypothesis, including extra information about the problem in the model itself, rather than in the training data.  4, fiche 1, Anglais, - inductive%20bias

Fiche 1,  Justifications,  Français

Record number: 1, Textual support number: 1 DEF

Hypothèse ou information a priori qu'un algorithme d'apprentissage utilise pour prédire la sortie d'une donnée d'entrée qu'il n'a jamais rencontrée.  3, fiche 1, Français, - biais%20inductif

Fiche 1,  Justifications,  Espagnol

Fiche 2,  Justifications,  Anglais

Record number: 2, Textual support number: 1 CONT

The problem then arises of choosing the most desirable inductive assertion(s). In making such a choice, one must take into consideration the aspects of the particular inductive learning problem; therefore the definition of a "preference criterion" for selecting a hypothesis is part of the problem background knowledge.  1, fiche 2, Anglais, - preference%20criterion

Record number: 2, Textual support number: 1 OBS

Also applied to the classification of entities in programming.  2, fiche 2, Anglais, - preference%20criterion

Fiche 2,  Justifications,  Français

Record number: 2, Textual support number: 1 DEF

(...) ensemble des propriétés caractérisant les généralisations intéressantes vis à vis de l'application considérée (...).  2, fiche 2, Français, - crit%C3%A8re%20de%20pr%C3%A9f%C3%A9rence

Record number: 2, Textual support number: 1 CONT

(...) pour restreindre l'espace des généralisations possibles (...) certains ont également proposé des critères de préférence basés sur des mesures d'utilité d'efficacité.  3, fiche 2, Français, - crit%C3%A8re%20de%20pr%C3%A9f%C3%A9rence

Fiche 2,  Justifications,  Espagnol

Fiche 3,  Justifications,  Anglais

Record number: 3, Textual support number: 1 CONT

To make the search computationally tractable, DEBUGGY assumes that the simple bugs that interact to form multiple ones can also be detected individually. This means that there exists at least one problem in which either the simple bugs appear in isolation or the complex bug is consistent with their individual manifestations. On this assumption, the generate-and-test method is first used to construct an initial hypothesis space. Every primitive bug is tested, plus a few very common combinations, and only those for which there is evidence in the data are kept. The initial hypothesis space is further reduced by eliminations of certain overlaps in coverage, with preference for multiple over single evidence. Next, the bugs in this initial hypothesis space are combined, and every combination that explains more than its parts is retained in the set of hypotheses.  1, fiche 3, Anglais, - initial%20hypothesis%20space

Fiche 3,  Justifications,  Français

Record number: 3, Textual support number: 1 OBS

Voir espace de recherche, hypothèse initiale.  1, fiche 3, Français, - espace%20d%27hypoth%C3%A8se%20initiale

Fiche 3,  Justifications,  Espagnol