Discriminative models, not discriminative training

Tom Minka

Abstract

The machine learning technique sometimes called "discriminative training" can be interpreted as ordinary training of a "discriminative model." This interpretation provides a better understanding of what discriminative training is really doing, and how it can be extended to partially-labeled data.

Details

Publication typeTechReport
NumberMSR-TR-2005-144
Pages1
InstitutionMicrosoft Research
> Publications > Discriminative models, not discriminative training