Uncertainty quantification (UQ) assists with the reliable and robust modeling of complex stochastic dynamics. SSL effortlessly develops novel information-theoretic tools to address the challenges concerning the feasibility and efficiency of UQ in stochastic systems.
SLL researchers develop methods and algorithms for learning the structure and the parameters of dynamical system from temporal measurements. Applications include but not limited to biological networks of proteins, neuroscience, physics and finance.
Generative models based on Deep Neural Nets have shown unprecedented capabilities in sampling data from complex but unknown distributions. SLL is working in devising new algorithms for training generative models and particularly Generative Adversarial Networks (GANs).
SLL researchers presented new algorithmic tools for investigating associations between a group of curves, considered as functional predictors, and a categorical outcome. The proposed schemes identify significant segments of the functional support for each covariate, which are combined with stepwise procedures to maximize classification accuracy.
Members of the SLL group conduct research on Bayesian estimation of linear and nonlinear parametric time series models (ARIMA, Threshold Autoregressions, Smooth Transition Autoregressions). Particular focus is devoted to the forecasting performaces of different types of shrinkage priors and stepwise model building algorithms, based on Kullback-Leibler distances.
SLL researchers developed new algorithms for parameter estimation and inference on spatio-temporal models with spatially varying coefficients. The proposed methodology combines Moran eigenvector filtering with sure independence screening and fast de-biased lasso-type estimators.