This research topic includes:
- Kernel-based identification of dynamical systems
- Neural Networks with stability guarantees for the identification of dynamical systems
- Optimization and learning algorithms
1. Kernel-based identification of dynamical systems
In the last decade, a cross-fertilization began between the System Identification and the Statistical Learning communities. This led firstly to the introduction of regularization techniques in system identification, and, more recently, to the application of kernel methods to dynamical systems learning. This line of research aims to leverage methodologies born under the machine learning light and employing them for system identification cause.
Methodological extensions we investigated in this area include:
- Nonparametric kernel methods for system identification using semi-supervised techniques
- Nonparametric kernel methods for continuous-time system identification
Selected publications
M Scandella, A Moreschini, T Parisini
Kernel-Based Continuous-Time System Identification: A Parametric Approximation
62nd IEEE Conference on Decision and Control (CDC), 1492-1497, 2023
M Scandella, M Bin, T Parisini
Kernel-Based Identification of Incrementally Input-to-State Stable Nonlinear Systems
22th IFAC World Congress, 56 (2), 5127-5132, 2023
P. Boni, M. Mazzoleni, M. Scandellaa, F. Previdi
A graph learning approach for kernel-based system identification with manifold regularization.
Journal of Franklin Institute, 2025.
M. Mazzoleni, A. Chiuso, M. Scandella, S. Formentin, F. Previdi
Kernel-based system identification with manifold regularization: a Bayesian perspective.
Automatica, vol. 142, pp. 110419, 2022, ISSN: 0005-1098. DOI: 10.1016/j.automatica.2022.110419.
M. Scandella, M. Mazzoleni, S. Formentin, F. Previdi
Kernel-based identification of asymptotically stable continuous-time linear dynamical systems.
International Journal of Control, 2020. DOI: 10.1080/00207179.2020.1868580.
S. Formentin, M. Mazzoleni, M. Scandella, F. Previdi
Nonlinear system identification via data augmentation.
Systems & Control Letters, 2019. DOI: 10.1016/j.sysconle.2019.04.004. ISSN: 0167-6911.
M. Scandella, M. Mazzoleni, S. Formentin, F. Previdi
A note on the numerical solutions of kernel-based learning problems.
IEEE Transactions on Automatic Control, 2019. DOI: 10.1109/TAC.2020.2989769.
2. Neural Networks with stability guarantees for the identification of dynamical systems
This line of research aims to provide stability guarantees for the training of neural networks model for nonlinear dynamical systems.
Methodological extensions we investigated in this area include:
- Input-State-Stability (ISS) and incremental ISS of Minimum Gated Unit
- Meta-learning approaches
Selected publications
3. Optimization and learning algorithms
This line of research aims to provide novel or improved optimization algorithm for the identification of dynamical systems or closed loop control policies.
Methodological extensions we investigated in this area include:
- Preference-based and black-box optimization
- Functional learning in Reproducing Kernel Hilbert Spaces
Selected publications
A Moreschini, M Scandella, T Parisini
Non-Convex Learning with Guaranteed Convergence: Perspectives on Stochastic Optimal Control
IEEE 63rd Conference on Decision and Control (CDC), 6002-6009, 2024
D. Previtali, M. Mazzoleni, A. Ferramosca, F. Previdi
GLISp-r: A preference-based optimization algorithm with convergence guarantees.
Computational Optimization and Applications, vol. 86, pp. 383-420, 2023. DOI: 10.1007/s10589-023-00491-2
