Nonparametric maximum likelihood estimation for large random graphs with latent data
Sylvain Le Corff, Université Paris-Sud, will give a lecture, organised by CMLA, focused on the estimation of the distribution of unobserved nodes in large random graphs from the observation of very few edges.
Le 17/05/2018 :
12:00 à 13:00
Ajouter à mon agenda
2024-04-24 16:59:21
2024-04-24 16:59:21
Nonparametric maximum likelihood estimation for large random graphs with latent data
Sylvain Le Corff, Université Paris-Sud, will give a lecture, organised by CMLA, focused on the estimation of the distribution of unobserved nodes in large random graphs from the observation of very few edges.
Bâtiment Cournet, Salle C103
ENS-PARIS-SACLAY
webmaster@ens-paris-saclay.fr
Europe/Paris
public
These graphs naturally model tournaments involving a large number of players (the nodes) where the ability to win of each player is unknown.
The players are only partially observed through discrete valued scores (edges) describing the results of contests between players. In this very sparse setting, we present the first nonasymptotic risk bounds for maximum likelihood estimators (MLE) of the unknown distribution of the nodes.
The proof relies on the construction of a graphical model encoding conditional dependencies that is extremely efficient to study n-regular graphs obtained using a round-robin scheduling.
This graphical model allows to prove geometric loss of memory properties and deduce the asymptotic behavior of the likelihood function.
Following a classical construction in learning theory, the asymptotic likelihood is used to define a measure of performance for the MLE. Risk bounds for the MLE are finally obtained by subgaussian deviation results derived from concentration inequalities for Markov chains applied to our graphical model. Bayesian posterior concentration rates for the unknown distribution are also proposed.
This graphical model allows to prove geometric loss of memory properties and deduce the asymptotic behavior of the likelihood function.
Following a classical construction in learning theory, the asymptotic likelihood is used to define a measure of performance for the MLE. Risk bounds for the MLE are finally obtained by subgaussian deviation results derived from concentration inequalities for Markov chains applied to our graphical model. Bayesian posterior concentration rates for the unknown distribution are also proposed.