CARE' Algo

Algo Carè has years of research experience in randomised methods for optimisation and data-driven optimisation. His expertise includes statistical learning and probability theory at large. He holds degrees in Computer Science and Control Systems. Almost all of his research interests are related to the fundamental generalisation issue, that is, to the question about when and how it is possible to get reliable information on an indefinitely large set of possible situations (experiments, input-output data, scenarios, etc.) based on a finite sample of them. Since the beginning of his research activity he has been intensively collaborating with two of the leading contributors to the theory of Scenario Approach, Professor Marco C. Campi (University of Brescia, Italy) and Professor Simone Garatti (Politecnico di Milano, Italy). In the Scenario Approach, one aims at finding a good solution to an optimisation problem with an infinite amount of constraints based on a finite number of them. His main contributions to the Scenario Approach are new algorithms that exploit dimension reduction and sparsity to optimise high dimensional uncertain problems in a provably reliable way; a theorem on invariant properties of data-based solutions to min-max problems; an extension of the Scenario Approach to the somewhat classic but theoretically challenging setting of least squares estimation methods. He also contributed on applications of the Scenario Approach, ranging from water management systems to power systems. The problem of learning in a reliable way from experimental data is also at the core of Dr Carè’s contributions to machine learning, which are focused on the classification problem and its applications in medicine (ventricular fibrillation in particular). Dr. Carè's research in system identification aim at evaluating the uncertainties in the models of dynamical systems that are built from a finite sample of input-output signals. In particular, his work is focused on finite-sample methods such as LSCR (Leave-out Sign-dominant Correlation Regions), SPS (Sign-Perturbed Sums) and the recently introduced SPCR (Sign-Perturbed Correlation Regions).