Séminaire Datashape
Sampling with Stein variational gradient descent
19
May 2022
May 2022
Intervenant : | Thomas Bonis |
Institution : | Laboratoire d'Analyse et de Mathématiques Appliquées de l'Université Gustave Eiffel |
Heure : | 15h45 |
Lieu : | 3L15 |
Many methods in Bayesian statistics require approximating intractable integrals. After presenting the usual methods used to deal with this issue, namely variational inference and Monte-Carlo approaches, along with their respective shortcomings, I will introduce a novel algorithm called Stein variational gradient descent and present known theoretical results regarding this algorithm.