Research activity
I am a research scientist with a background in scientific computing, and the core of my research activity falls in the areas of Numerical Optimization and Machine Learning.
Following the studies in my Master’s Thesis [A], my first works have been concerned with quadratic programming problems arising in the context of Machine Learning. More specifically, I have been studying fast approximation algorithms for the Support Vector Machine training problem based on classical Frank-Wolfe type optimization methods. Some results in this sense can be found in [A,1,2,3]. In [6], these algorithms are presented within a more general framework, and a new and more effective variant of the Frank-Wolfe method is presented and analyzed.
During the period I spent as a PhD student, I explored the possibility of enhancing classical Direct Search algorithms by embedding them in a multilevel optimization paradigm. Inspired by classical Multigrid methods for linear and nonlinear systems of equations, multilevel schemes can greatly enhance the performance of an underlying optimization solver. As a result, traditional limitations on the problem size and the accuracy of the solution typically found in Direct Search optimization can be overcome [B,4,5]. Applications range from classical discretized optimization problems arising from the discretization of an infinite-dimensional functional to the reconstruction of blurred and noisy images with a Total Variation deblurring model [B].
Later, I have been working for 2.5 years as a postdoctoral research scientists in the ESAT-STADIUS group at KU Leuven, in the context of the ERC Advanced Grant A-DATADRIVE-B on Advanced Data-Driven Black-box Modelling (PI: Prof. Johan A. K. Suykens). My research there focused on the study of high-performance optimization algorithms specifically targeted to Machine Learning and Data Mining. Main applications included classical Support Vector Machines [8,9], LASSO regression [10] and matrix recovery problems [7]. The use of modern computational resources (such as parallel and GPU computing) to maximize their performance was also part of my research interests [7].
I am currently a JSPS Postdoctoral Fellow at the RIKEN Brain Science Institute in Japan. The goal of my research project is to develop cutting-edge Machine Learning based models to investigate the mechanics of the neural computations lying at the core of perception and decision-making.