|
FEDERICO CARLI
The R Package stagedtrees for Structural Learning of Stratified Staged Trees
stagedtrees is an R package which includes several algorithms for learning the structure of staged trees and chain event graphs. Both score-based and distance-based algorithms are implemented, as well as various functionalities to plot the models and perform inference. CLAUDIA DI CATERINA Location-adjusted Wald statistics for scalar parameters The location of the Wald statistic is adjusted in order to obtain significant improvements in inferential performance with only small implementation and computational overhead. Analytical and numerical evidence for the adoption of the location-adjusted statistic is provided for prominent modelling frameworks. DAVIDE GARBARINO Scale-free property in the World Wide Web Nowadays many random graph models have been proposed and ana- lyzed, referring to certain common features observed in large-scale real- word graphs such as the 'web graph', whose nodes are web-pages, with a directed link between two web pages. One of the main observation is that the graphs are scale-free, naively meaning that the distribution of the vertex degrees follows a Pareto distribution rather than the Poisson distribution of the classical Erdös-Renyi random graph model. Nevertheless, few rigorous statistical procedures have been adopted in or- der to statistically prove the ubiquitousness of the scale-free property in real-world networks. In particular, in this talk I am going to present a procedure introduced by Clauset et al., focusing on a particular kind of data I am interested in : institutional websites and peripheral editorial websites. MANUELE LEONELLI Diagonal Distributions We introduce diagonal distributions as an extension of marginal distributions. The main diagonal is studied in detail, which consists of a mean-constrained univariate distribution function on [0; 1] that summarizes key features on the dependence structure of a random vector, whose variance connects with Spearman's rho, and whose mass at the endpoints 0 and 1 offers insights on the strength of tail dependence. Mean-constrained histograms and mean-constrained kernel-based methods are developed so to learn about the main diagonal density from data. Simulations show that proposed estimators accurately recover the true main diagonal under a variety of simulation scenarios. An application is discussed illustrating how diagonal densities can be used so to contrast the diversication of a portfolio based on FAANG stocks against one based on crypto-assets. The talk is based on joint work with Miguel de Carvalho, Raphael Huser and Rodrigo Rubio. MARTA NAI RUSCONE Model-based clustering through copula Finite mixtures are applied to perform model-based clustering of multivariate data. Existing models are not flexible enough for modelling the dependence of multivariate data since they rely on potentially undesirable correlation restrictions and strict assumptions on the marginal distribution to be computationally tractable. We discuss a model-based clustering method via copula to understand the complex and hidden dependence patterns in correlated multivariate data. We also propose a parsimonious version of model-based clustering method via R-vine copula to alleviate the computational burden and the risk of overfitting. One of the advantages of this approach is that it accounts for the tail asymmetry of the data by using blocks of asymmetric bivariate copulas. We use simulated and real datasets to illustrate the proposed methodology. ELENA PESCE Large Datasets, Bias and Model Oriented Optimal Design of Experiments Statistical methods have been developed for the analysis of Big Datasets, which use the full available data. In contrast other authors argue on the advantages of inference statements based on a well-chosen subset of the big dataset. We review recent literature that proposes to adapt ideas from classical model based optimal design of experiments to problems of data selection of large datasets. Special attention is given to bias reduction and to protection against confounders. Some new results are presented. Theoretical and computational comparisons are made. GHERARDO VARANDO Graphical continuous Lyapunov models We introduce a new graphical model via the linear Lyapunov equation of a covariance matrix. This model class parametrizes equilibrium covariances for stochastic processes. We show how the model class behaves under marginalization and introduce a method for structure learning via l1-penalized loss minimization. |
|