Context-based decision may help for interactive learning and domain adaptation
Résumé
Deep networks trained in a supervised way can achieve impressive performance in the initial distribution while performing poorly in close-but-different distributions. Although there is already a large literature on domain adaptation and interactive learning aiming to deal with such performance drop, the use of few labeled data from target distribution as context has not been widely studied. This technical report points out that using a context-like framework can slightly improve the performance of the standard deep network under moderate domain shift.
Domaines
Machine Learning [stat.ML]Origine | Fichiers produits par l'(les) auteur(s) |
---|