|
Statut |
Confirmé |
|
Série |
SEM-LPTM-UCP |
|
Domaines |
CoRR.LG |
|
Date |
Mardi 4 Juin 2019 |
|
Heure |
14:00 |
|
Institut |
LPTM |
|
Salle |
4.13 St Martin II |
|
Nom de l'orateur |
Barra |
|
Prenom de l'orateur |
Adriano |
|
Addresse email de l'orateur |
|
|
Institution de l'orateur |
Dpt. di Matematica & Fisica "Ennio De Giorgi", Università del Salento, Italie |
|
Titre |
A journey in the statistical mechanical approach to neural networks: some novel results' |
|
Résumé |
In this talk, provided a streamlined introduction to statistical
learning and related information retrieval achieved by associative
neural networks, I will deepen some recent progresses via
their statistical mechanical formalization. I will focus on the
nature of the information to store (i.e. analog vs digital patterns)
and on the role of dilution in the underlying graphs where the neurons are hosted (particularly relating network 's dilution to spontaneous
multi-tasking capabilities the net experiences as a whole). Finally, I
will discuss how by 'letting these machines sleep' it is possible to
sensibly increase their performances, much as their real (i.e.
biological) counterparts. |
|
Numéro de preprint arXiv |
|
|
Commentaires |
Ref: https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.109.268101 |
|
Fichiers attachés |
|
