arXiv subjects

Mario Sznaier

Publications and source records attributed to Mario Sznaier.

At least 19 recordsLinked to original sources

arXiv2026

Frequency Response Identification of Low-Order Systems: Finite-Sample Analysis

This paper proposes a frequency-domain estimator for low-order systems from repeated noisy measurements. The estimator minimizes a quadratic data-fitting term regularized by the nuclear norm of a Loewner matrix, subject to a convex stability constraint enforced via a semidefinite program. We prove a finite-sample error bound at the sampled frequencies and extend it to all frequencies through rational interpolation. The bound characterizes the dependence on the number of repeated experiments, number of frequency points, system order, and noise level. Numerical experiments on SISO and MIMO systems demonstrate the low-order-promoting effect of the method and validate the predicted scaling laws.

eess.SY↗

arXiv2026

Time-Varying Perturbations of Contractive Systems With an Application to Safe Stabilization

Perturbation theory for asymptotically stable systems has received sustained attention, leading to many criteria that preserve stability under uncertainty. A limitation of classical stability analysis, however, is that it is inherently equilibrium-dependent. Contractive systems, by contrast, are defined independently of any equilibrium and admit a rich set of robustness bounds. In this letter, we develop new time-varying perturbation conditions for contractive dynamics that preserve incremental exponential stability of the perturbed system and, in specific regimes, guarantee asymptotic convergence to solutions of the nominal dynamics. As an application, we propose a safety filter for steering a nominally contractive system to a desired equilibrium while avoiding an unsafe set, and we present an example illustrating advantages of the proposed method over existing approaches.

math.OC↗

arXiv2026

Randomized Atomic Feature Models for Physics-Informed Identification of Dynamic Systems

We present a physics-informed framework for system identification based on randomized stable atomic features. Impulse responses are represented as random superpositions of stable atoms, namely damped complex exponentials associated with poles sampled inside a prescribed disk. Identification is then cast as a convex regularized least-squares problem with optional linear, second-order-cone, and KYP constraints. The approach generalizes random Fourier and random Laplace features to the damped, nonstationary regime relevant to engineering systems while retaining modal interpretability and scalable finite-dimensional computation. The main analytic point is an operator-theoretic Disk-Bochner viewpoint: positive measures over stable poles generate positive-definite kernels with a radius-dependent shift defect, while a converse scalar disk moment representation for an arbitrary kernel is characterized by subnormality of the canonical shift. We prove this statement, establish an RKHS-to-l1 embedding, show that sampled poles induce a valid finite atomic gauge, discuss random-feature convergence, and state sparse-recovery guarantees conditionally on the restricted-eigenvalue properties of the realized disk-Vandermonde or input-output design matrix. We also connect the normalized transfer function problem to Nevanlinna-Pick interpolation and LFT set-membership. The framework directly encodes stability margins, modal localization, DC-gain bounds, monotonicity, passivity, relative degree, settling-time targets, and time/frequency-domain error bounds. Numerical comparisons illustrate how physically meaningful priors can compensate for poor excitation and improve constrained impulse-response recovery in an under-informative data setting.

eess.SY↗

arXiv2026

Unsafe Probabilities and Risk Contours for Stochastic Processes using Convex Optimization

This paper proposes an algorithm to calculate the maximal probability of unsafety with respect to trajectories of a stochastic process and a hazard set. The unsafe probability estimation problem is cast as a primal-dual pair of infinite-dimensional linear programs in occupation measures and continuous functions. This convex relaxation is nonconservative (to the true probability of unsafety) under compactness and regularity conditions in dynamics. The continuous-function linear program is linked to existing probability-certifying barrier certificates of safety. Risk contours for initial conditions of the stochastic process may be generated by suitably modifying the objective of the continuous-function program, forming an interpretable and visual representation of stochastic safety for test initial conditions. All infinite-dimensional linear programs are truncated to finite dimension by the Moment-Sum-of-Squares hierarchy of semidefinite programs. Unsafe-probability estimation and risk contours are generated for example stochastic processes.

math.OC↗

arXiv2026

On incremental and semi-global exponential stability of gradient flows satisfying generalized Łojasiewicz inequalities

The Łojasiewicz inequality characterizes objective-value convergence along gradient flows and, in special cases, yields exponential decay of the cost. However, such results do not directly give rates of convergence in the state. In this paper, we use contraction theory to derive state-space guarantees for gradient systems satisfying generalized Łojasiewicz inequalities. We first show that, when the objective has a unique strongly convex minimizer, the generalized Łojasiewicz inequality implies semi-global exponential stability; on arbitrary compact subsets, this yields exponential stability. We then give two curvature-based sufficient conditions, together with constraints on the Łojasiewicz rate, under which the nonconvex gradient flow is globally incrementally exponentially stable.

math.OC↗

arXiv2025

Generalization Error Analysis for Selective State-Space Models Through the Lens of Attention

State-space models (SSMs) have recently emerged as a compelling alternative to Transformers for sequence modeling tasks. This paper presents a theoretical generalization analysis of selective SSMs, the core architectural component behind the Mamba model. We derive a novel covering number-based generalization bound for selective SSMs, building upon recent theoretical advances in the analysis of Transformer models. Using this result, we analyze how the spectral abscissa of the continuous-time state matrix influences the model's stability during training and its ability to generalize across sequence lengths. We empirically validate our findings on a synthetic majority task, the IMDb sentiment classification benchmark, and the ListOps task, demonstrating how our theoretical insights translate into practical model behavior.

cs.LG↗

arXiv2025

Real-Time Adaptive Motion Planning via Point Cloud-Guided, Energy-Based Diffusion and Potential Fields

Motivated by the problem of pursuit-evasion, we present a motion planning framework that combines energy-based diffusion models with artificial potential fields for robust real time trajectory generation in complex environments. Our approach processes obstacle information directly from point clouds, enabling efficient planning without requiring complete geometric representations. The framework employs classifier-free guidance training and integrates local potential fields during sampling to enhance obstacle avoidance. In dynamic scenarios, the system generates initial trajectories using the diffusion model and continuously refines them through potential field-based adaptation, demonstrating effective performance in pursuit-evasion scenarios with partial pursuer observability.

cs.RO↗

arXiv2025

Robust Data-Driven Receding-Horizon Control for LQR with Input Constraints

This letter presents a robust data-driven receding-horizon control framework for the discrete time linear quadratic regulator (LQR) with input constraints. Unlike existing data-driven approaches that design a controller from initial data and apply it unchanged throughout the trajectory, our method exploits all available execution data in a receding-horizon manner, thereby capturing additional information about the unknown system and enabling less conservative performance. Prior data-driven LQR and model predictive control methods largely rely on Willem's fundamental lemma, which requires noise-free data, or use regularization to address disturbances, offering only practical stability guarantees. In contrast, the proposed approach extends semidefinite program formulations for the data-driven LQR to incorporate input constraints and leverages duality to provide formal robust stability guarantees. Simulation results demonstrate the effectiveness of the method.

math.OC↗

arXiv2025

Robust Data-Driven Receding Horizon Control

This paper presents a data-driven receding horizon control framework for discrete-time linear systems that guarantees robust performance in the presence of bounded disturbances. Unlike the majority of existing data-driven predictive control methods, which rely on Willem's fundamental lemma, the proposed method enforces set-membership constraints for data-driven control and utilizes execution data to iteratively refine a set of compatible systems online. Numerical results demonstrate that the proposed receding horizon framework achieves better contractivity for the unknown system compared with regular data-driven control approaches.

math.OC↗

arXiv2025

Modeling Adaptive Tracking of Predictable Stimuli in Electric Fish

The weakly electric fish \emph{Eigenmannia virescens} naturally swims back and forth to stay within a moving refuge, tracking its motion using visual and electrosensory feedback. Previous experiments show that when the refuge oscillates as a low-frequency sinusoid (below about 0.5 Hz), the tracking is nearly perfect, but phase lag increases and gain decreases at higher frequencies. Here, we model this nonlinear behavior as an adaptive internal model principle (IMP) system. Specifically, an adaptive state estimator identifies the \emph{a priori} unknown frequency, and feeds this parameter estimate into a closed-loop IMP-based system built around a lightly damped harmonic oscillator. We prove that the closed-loop tracking error of the IMP-based system, where the online adaptive frequency estimate is used as a surrogate for the unknown frequency, converges exponentially to that of an ideal control system with perfect information about the stimulus. Simulations further show that our model reproduces the fish refuge tracking Bode plot across a wide frequency range. These results establish the theoretical validity of combining the IMP with an adaptive identification process and provide a basic framework in adaptive sensorimotor control.

eess.SY↗

arXiv2025

SAIF: Sparse Adversarial and Imperceptible Attack Framework

Adversarial attacks hamper the decision-making ability of neural networks by perturbing the input signal. The addition of calculated small distortion to images, for instance, can deceive a well-trained image classification network. In this work, we propose a novel attack technique called Sparse Adversarial and Interpretable Attack Framework (SAIF). Specifically, we design imperceptible attacks that contain low-magnitude perturbations at a small number of pixels and leverage these sparse attacks to reveal the vulnerability of classifiers. We use the Frank-Wolfe (conditional gradient) algorithm to simultaneously optimize the attack perturbations for bounded magnitude and sparsity with $O(1/\sqrt{T})$ convergence. Empirical results show that SAIF computes highly imperceptible and interpretable adversarial examples, and outperforms state-of-the-art sparse attack methods on the ImageNet dataset.

cs.CV↗

arXiv2025

Convex Data-Driven Contraction With Riemannian Metrics

The growing complexity of dynamical systems and advances in data collection necessitates robust data-driven control strategies without explicit system identification and robust synthesis. Data-driven stability has been explored in linear and nonlinear systems, often by turning the problem into a linear or positive semidefinite program. This paper focuses on a new emerging property called contractivity, which refers to the exponential convergence of all system trajectories toward each other under a specified metric. Data-driven closed loop contractivity has been studied for the case of the 2-norm and assuming nonlinearities are Lipschitz bounded in subsets of n dimensional euclidean space. We extend the analysis by considering Riemannian metrics for polynomial dynamics. The key to our derivation is to leverage the convex criteria for closed-loop contraction and duality results to efficiently check infinite dimensional membership constraints. Numerical examples demonstrate the effectiveness of the proposed method for both linear and nonlinear systems, highlighting its potential for robust data-driven contraction.

math.OC↗

arXiv2025

Safe Control for Pursuit-Evasion with Density Functions

This letter presents a density function based safe control synthesis framework for the pursuit-evasion problem. We extend safety analysis to dynamic unsafe sets by formulating a reach-avoid type pursuit-evasion differential game as a robust safe control problem. Using density functions and semi-algebraic set definitions, we derive sufficient conditions for weak eventuality and evasion, reformulating the problem into a convex sum-of-squares program solvable via standard semidefinite programming solvers. This approach avoids the computational complexity of solving the Hamilton-Jacobi-Isaacs partial differential equation, offering a scalable and efficient framework. Numerical simulations demonstrate the efficacy of the proposed method.

math.OC↗

arXiv2025

Challenges in Model Agnostic Controller Learning for Unstable Systems

Model agnostic controller learning, for instance by direct policy optimization, has been the object of renewed attention lately, since it avoids a computationally expensive system identification step. Indeed, direct policy search has been empirically shown to lead to optimal controllers in a number of cases of practical importance. However, to date, these empirical results have not been backed up with a comprehensive theoretical analysis for general problems. In this paper we use a simple example to show that direct policy optimization is not directly generalizable to other seemingly simple problems. In such cases, direct optimization of a performance index can lead to unstable pole/zero cancellations, resulting in the loss of internal stability and unbounded outputs in response to arbitrarily small perturbations. We conclude the paper by analyzing several alternatives to avoid this phenomenon, suggesting some new directions in direct control policy optimization.

math.OC↗

arXiv2025

3D-HGS: 3D Half-Gaussian Splatting

Photo-realistic image rendering from 3D scene reconstruction has advanced significantly with neural rendering techniques. Among these, 3D Gaussian Splatting (3D-GS) outperforms Neural Radiance Fields (NeRFs) in quality and speed but struggles with shape and color discontinuities. We propose 3D Half-Gaussian (3D-HGS) kernels as a plug-and-play solution to address these limitations. Our experiments show that 3D-HGS enhances existing 3D-GS methods, achieving state-of-the-art rendering quality without compromising speed.

cs.CV↗

arXiv2025

A Hybrid Framework for Efficient Koopman Operator Learning

Koopman analysis of a general dynamics system provides a linear Koopman operator and an embedded eigenfunction space, enabling the application of standard techniques from linear analysis. However, in practice, deriving exact operators and mappings for the observable space is intractable, and deriving an approximation or expressive subset of these functions is challenging. Programmatic methods often rely on system-specific parameters and may scale poorly in both time and space, while learning-based approaches depend heavily on difficult-to-know hyperparameters, such as the dimension of the observable space. To address the limitations of both methods, we propose a hybrid framework that uses semidefinite programming to find a representation of the linear operator, then learns an approximate mapping into and out of the space that the operator propagates. This approach enables efficient learning of the operator and explicit mappings while reducing the need for specifying the unknown structure ahead of time.

eess.SY↗

arXiv2024

Quantifying the Safety of Trajectories using Peak-Minimizing Control

This work quantifies the safety of trajectories of a dynamical system by the perturbation intensity required to render a system unsafe (crash into the unsafe set). Computation of this measure of safety is posed as a peak-minimizing optimal control problem. Convergent lower bounds on the minimal peak value of controller effort are computed using polynomial optimization and the moment-Sum-of-Squares hierarchy. The crash-safety framework is extended towards data-driven safety analysis by measuring safety as the maximum amount of data corruption required to crash into the unsafe set.

math.OC↗

arXiv2024

Data-Driven Superstabilizing Control under Quadratically-Bounded Errors-in-Variables Noise

The Error-in-Variables model of system identification/control involves nontrivial input and measurement corruption of observed data, resulting in generically nonconvex optimization problems. This paper performs full-state-feedback stabilizing control of all discrete-time linear systems that are consistent with observed data for which the input and measurement noise obey quadratic bounds. Instances of such quadratic bounds include elementwise norm bounds (at each time sample), energy bounds (across the entire signal), and chance constraints arising from (sub)gaussian noise. Superstabilizing controllers are generated through the solution of a sum-of-squares hierarchy of semidefinite programs. A theorem of alternatives is employed to eliminate the input and measurement noise process, thus improving tractability.

math.OC↗