I am a Postdoctoral Researcher within the research group System Identification and Control (SIC), Department of Control and Computer Engineering (DAUIN), Politecnico di Torino, Italy.
I am a teaching assistant in the “Laboratory of Robust Identification and Control” course in the Master's degree in Mechatronic Engineering, Politecnico di Torino.
I love playing chess; challenge me on chess.com! Username: Enimos.
Ph.D Thesis Defense
On Friday, May 16th, I successfully defended my PhD thesis:
“Optimization Methods for Dynamical Systems Learning.”
My research developed new algorithms for constrained optimization using a control-theoretic approach. These are then applied to the system identification problem to avoid the vanishing gradient issue.
Throughout these three years, I’ve learned far more than I ever imagined, not only about control, optimization, and learning but also about scientific rigor and the importance of clear presentation and writing. I’ve also learned to approach challenges with curiosity and resilience and to see setbacks as opportunities for growth.
Heartfelt thanks to my amazing supervisors, Diego Regruto and Sophie M. Fosson. This achievement would not have been possible without their guidance. Thanks to my friends and colleagues at Laboratory #11 – DAUIN for creating such a nice work environment. Special thanks to my partner Jennifer Fera and my family, whose support helped me maintain a healthy work-life balance.
I also want to thank my defense committee, Carlo Novara, Emiliano Dall'Anese, Daniele Astolfi, Andrea Calimera, and Antonello Giannitrapani, for their truly inspiring comments.
New Paper Accepted!
🚀 Thrilled to announce our latest publication in the IEEE Transactions on Automatic Control
“A New Framework for Constrained Optimization via Feedback Control of Lagrange Multipliers”
🔗 Available on IEEE Xplore
🔗 Open access on ArXiv
This paper introduces a novel approach to solving constrained optimization problems by dynamically regulating the Lagrange multipliers. The proposed framework enables the development of new optimization algorithms using feedback controller design: any controller corresponds to a new algorithm. In the paper, we explore proportional-integral and feedback linearization to develop two new algorithms, and we analyze their properties.
📖 This research constitutes the core part of my PhD thesis work. I’m excited to build on these results and can’t wait to explore future developments at the intersection of control theory and optimization!
🙌 Heartfelt thanks to my co-authors and supervisors, Vito Cerone, Sophie M. Fosson, and Diego Regruto. Your guidance, expertise, and continuous support were essential in shaping this research.
My research interests are in the areas of
optimization algorithms;
dynamical systems theory;
system identification;
data-driven and robust control.
Optimization algorithms design
We address the problem of developing efficient optimization algorithms.
We address constrained optimization problems, proposing a framework based on feedback control theory. We define a fictitious plant where the Lagrange multipliers act as a control input used to regulate the constraints. Depending on the designed controller, the closed-loop dynamics enjoys different convergence properities and computational complexity.
System Identification
We address the problem of estimating the parameters of black-box and gray-box models, considering continuous and discrete time. Specifically, we study:
The set-membership identification problem, where the goal is to estimate bounds on each model's parameters. In this scenario, we focus on linear systems and provide a solution based on polynomial optimization.
The simulation error minimization problem. Considering general nonlinear input-output models, we propose a novel training algorithm unaffected by vanishing gradient, based on controlled multipliers optimization.
Data-driven controller design
We address the problem of directly designing feedback controllers from data without resorting to plant estimation/identification. We place particular emphasis on guaranteeing the feedback system's stability in the presence of finite data and bounded noise.