discriminative model

noun

A discriminative model is a type of machine learning model designed to predict the probability of a specific output given an input, rather than modeling the underlying data distribution.

Unlike generative models, which learn how data is created, discriminative models focus on decision boundaries. For example, a discriminative model trained on images might learn to classify them as ‘cat’ or ‘dog’ by analyzing pixel patterns.

For users, this means faster, more accurate predictions in tasks like image recognition or language translation. However, it requires large, labeled datasets and can struggle with rare or ambiguous cases. For instance, a discriminative model might correctly identify 95% of cats in images but fail on unusual breeds without enough training examples.

A discriminative model in a photo app automatically tags pictures of pets as 'cat' or 'dog' without needing to generate new images.

See also: generative model · Supervised learning

Découvrez nos autres services

ElpisIA, c’est tout un univers — explorez nos autres services.