v Piecewise nonlinear regression with data augmentation · Mirko Mazzoleni

Piecewise nonlinear regression with data augmentation

M. Mazzoleni, V. Breschi, S. Formentin

Abstract

Piecewise regression represents a powerful tool to derive accurate yet modular models describing complex phenomena or physical systems. This paper presents an approach for learning PieceWise NonLinear (PWNL) functions in both a supervised and semi-supervised setting. We further equip the proposed technique with a method for the automatic generation of additional unsupervised data, which are leveraged to improve the overall accuracy of the estimate. The performance of the proposed approach is preliminarily assessed on two simple simulation examples, where we show the benefits of using nonlinear local models and artificially generated unsupervised data. [Paper, Code]

Reference

M. Mazzoleni, V. Breschi, S. Formentin, "Piecewise nonlinear regression with data augmentation", in 19th IFAC Symposium on System Identification IFAC-PapersOnLine, vol. 54, Issue 7, 2021, pp. 421-426, ISSN 2405-8963, doi: 10.1016/j.ifacol.2021.08.396.

Bibtex

@article{MAZZOLENI2021421,
title = {Piecewise nonlinear regression with data augmentation},
journal = {IFAC-PapersOnLine},
volume = {54},
number = {7},
pages = {421-426},
year = {2021},
note = {19th IFAC Symposium on System Identification SYSID 2021},
issn = {2405-8963},
doi = {https://doi.org/10.1016/j.ifacol.2021.08.396}
}
Share this:

RELATED ARTICLES

Kernel-based system identification with manifold regularization: A Bayesian perspective
M. Mazzoleni, V. Breschi, S. Formentin
### This paper presents a nonparametric Bayesian interpretation of kernel-based function learning with manifold regularization. We show that manifold regularization corresponds to an additional likelihood term derived from noisy observations of the function gradient along the regressors graph. The hyperparameters of the method are estimated by ...
2022-04-30
Journal   
Nonparametric continuous-time identification of linear systems: theory, implementation and experimental results
M. Mazzoleni, V. Breschi, S. Formentin
### This paper presents an algorithm for continuous-time identification of linear dynamical systems using kernel methods. When the system is asymptotically stable, also the identified model is guaranteed to share such a property. The approach embeds the selection of the model complexity through optimization of the marginal likelihood of the dat...
2021-10-27
Conference   
Kernel-based identification of asymptotically stable continuous-time linear dynamical systems
M. Mazzoleni, V. Breschi, S. Formentin
In many engineering applications, continuous-time models are preferred to discrete-time ones, in that they provide good physical insight and can be derived also from non-uniformly sampled data. However, for such models, model selection is a hard task if no prior physical knowledge is given. In this paper, we propose a non-parametric approach to...
2021-01-29
Journal