Nos tutelles

CNRS Dauphine PSL *



8 juin 2017: 1 événement


  • Séminaires du Pôle 3 : "Sciences des données"

    Jeudi 8 juin 14:00-15:00 - Yamane Ikko - Tokyo University

    séminaire Pole 3 - Yamane Ikko - "Topics on multitask learning"

    Résumé : Title : Topics on multitask learning
    Abstract : Machine learning aims at developing techniques for automatically extracting useful general information from data using computers.
    In many practical applications, several related tasks have to be solved simultaneously. If enough data are available for each task, they can be solved independently but in the contrary, the performance of each task will be poor. On the other hand, although the tasks are related, they are usually sufficiently different so that a single common model fails at solving all the tasks at once.
    In this context, Multitask learning is a promising approach to extend conventional machine learning methods in order to handle such cases. Using the similarities among the tasks, it aims at solving the tasks while ensuring that the solutions are similar.
    After reviewing the basic concepts in multitask learning, this talk will focus on two recent applications in unsupervised learning, namely the multitask extensions of Principal Component Analysis and Density Gradient Estimation.

    Lieu : P303

    En savoir plus : Séminaires du Pôle 3 : "Sciences des données"