134,947 research outputs found
Consistent Sobolev regression via fuzzy systems with overlapping concepts
No full textIn this paper we propose a new nonparametric regression algorithm based on Fuzzy systems with overlapping concepts. We analyze its consistency properties, showing that it is capable to reconstruct an infinite-dimensional class of function when the size of the noisy dataset grows to infinity. Moreover, convergence to the target function is guaranteed in Sobolev norms so ensuring uniform convergence also for a certain number of derivatives. The
connection with Regularization Networks, Bayesian estimation and
Tychonov regularization is highlighted
Single-linkage clustering for optimal classification in piecewise affine regression
No full textWhen performing regression with piecewise affine maps, the most challenging task is to classify the data points, i.e. to correctly attribute a data point to the affine submodel that most likely generated it. In this paper, we consider a regression scheme similar to the one proposed in (Ferrari-Trecate et al., 2001,2003) that reduces the classification step to a clustering problem in presence of outliers. However, instead of the K-means procedure adopted in (Ferrari-Trecate et al., 2001,2003), we propose the use of single-linkage clustering that estimates automatically the number of submodels composing the piecewise affine map. Moreover we prove that, under mild assumptions on the data set, single-linkage clustering can guarantee optimal classification in presence of bounded noise.SCI-STI-GF
Fast spline smoothing via spectral factorization concepts
No full textWhen tuning the smoothness parameter of nonparametric regression splines, the evaluation of the so-called degrees of freedom is one of the most computer-intensive tasks. In the paper, a closed-form expression of the degrees of freedom is obtained for the case of cubic splines and equally spaced data when the number of data tends to infinity. State-space methods, Kalman filtering and spectral factorization techniques are used to prove that the asymptotic degrees of freedom are equal to the variance of a suitably defined stationary process. The closed-form expression opens the way to fast spline smoothing algorithms whose computational complexity is about one-half of standard methods (or even one-fourth under further approximations)
Zeros of continuous-time linear periodic systems
No full textZeros of continuous-time linear periodic systems are defined and their properties investigated. Under the assumption that the system has uniform relative degree, the zero-dynamics of the system is characterized and a closed-form expression of the blocking inputs is derived. This leads to the definition of zeros as unobservable characteristic exponents of a suitably defined periodic pair. The zeros of periodic linear systems satisfy blocking properties that generalize the well-known time-invariant case. Finally, an efficient computational scheme is provided that essentially amounts to solving an eigenvalue problem
Sobolev approximation by means of fuzzy systems with overlapping Gaussian concepts
No full textIn this paper the approximating capabilities of fuzzy systems with overlapping Gaussian concepts are considered. A new method for the computation of the system coefficients is provided, showing that it guarantees uniform approximation of the derivatives of the target function. Moreover, the connection with radial basis functions approximations is highlighte
Conditions of Optimal Classification for Piecewise Affine Regression
No full textWe consider regression problems with piecewise affine maps. In particular, we focus on the sub-problem of classifying the datapoints, i.e. correctly attributing a datapoint to the affine submodel that most likely generated it. Then, we analyze the regression algorithm proposed by Ferrari-Trecate et. al (2003) and show that, under suitable assumptions on the dataset and the weights of the classification procedure, optimal classification can be guaranteed in presence of bounded noise. We also relax such assumptions by introducing and characterizing the property of weakly optimal classification. Finally, by elaborating on these concepts, we propose a procedure for detecting, a posteriori, misclassified datapoints.SCI-STI-GF
A hybrid model predictive control scheme for containment and distributed sensing in multi-agent systems
No full textThis paper proposes a control scheme for distributed sensing using a leader/follower multi-agent architecture. The control objective is to make a group of mobile agents cover and sense a sequence of regions of interest. More specifically, when the leaders reach a new target region, they stop until the followers have performed a sensing task. Furthermore, the followers must be contained inside the convex-hull of the leaders’ positions during the motion. Key features of our method, that combines hybrid control with Model Predictive Control (MPC) techniques, are the possibility to take into account input constraints in order to plan the sensing maneuver and the ability of the followers to detect containment violations by simple computation based on the available information about the leaders’ positions
Deuxiéme Duo pour harpe et Piano Ou pour deux Pianos : Opera XX / Par G. G. Ferrari
No full textDEUXIÉME DUO POUR HARPE ET PIANO OU POUR DEUX PIANOS : OPERA XX / PAR G. G. FERRARI
Deuxiéme Duo pour harpe et Piano Ou pour deux Pianos : Opera XX / Par G. G. Ferrari (1)
Titelseite (1)
Verlagsanzeigen (2)
Noten (3
Design of plug-and-play model predictive control: an approach based on linear programming
No full textIn this paper we consider a linear system represented by subsystems coupled through states and propose a distributed control scheme for guaranteeing asymptotic stability and satisfaction of constraints on system inputs and states. Our design procedure enables Plug-and-Play (PnP) operations, meaning that (i) the addition or removal of subsystems triggers the synthesis of local controllers associated to successors to the subsystem only and (ii) the synthesis of a local controller for a subsystem requires information only from predecessors of the subsystem and it can be performed using only local computational resources. Our method, that is based on Model Predictive Control (MPC) advances the PnP design procedure proposed in [1] in several directions. Notably, we show how critical steps in the design of a local controller can be solved through linear programming
- …
