Raffaele Iervolino

This paper develops a behavioral data-driven framework for stability analysis and stabilization of autoregressive systems whose coefficients depend affinely on an unknown, unmeasured, and constant parameter vector. The affine parameter directions are known, while no prior bounds on the parameter values are assumed. From noisy input–output data, a quadratic residual description yields an explicit quadratic characterization of the complete set of compatible parameter values. For energy bounded residuals, a rank condition characterizes parametric informativity, while a strict consistency condition ensures, together with full parameter rank, that the compatible set is a nonempty bounded ellipsoid with nonempty interior. The resulting structured data compatible AR family is shown to coincide exactly with the intersection of the corresponding unstructured compatible family and the prescribed affine model set; equivalently, its parameter space QMI is obtained as the affine pullback of the unstructured coefficient space QMI. Quadratic open loop stability and stabilization are then formulated over the complete compatible parameter set. Exact robust conditions and finite dimensional sufficient LMI certificates are derived, including the synthesis of a single parameter independent dynamic behavioral controller. A numerical example shows that exploiting the affine structure can significantly reduce conservatism relative to an unstructured data-driven description.

Federated learning (FL) has emerged as a powerful paradigm for decentralized machine learning, enabling collaborative model training across diverse clients without sharing raw data. However, traditional FL approaches often face limitations in scalability and efficiency due to their reliance on synchronous client updates, which can result in significant delays and increased communication overhead, particularly in heterogeneous and dynamic environments. To address these challenges in this article, we propose an asynchronous FL (AFL) algorithm, which allows clients to update the global model independently and asynchronously. Our key contributions include a comprehensive convergence analysis of AFL in the presence of client delays and model staleness. By leveraging martingale difference sequence theory and variance bounds, we ensure robust convergence despite asynchronous updates. Assuming strongly convex local objective functions, we establish bounds on gradient variance under random client sampling and derive a recursion formula quantifying the impact of client delays on convergence. Furthermore, we demonstrate the practical applicability of the AFL algorithm by training decentralized linear regression and support vector machine (SVM)-based classifiers and compare its results with synchronous FL algorithm to effectively handle non-IID data distributed among clients. The proposed AFL algorithm addresses key limitations of traditional FL methods, such as inefficiency due to global synchronization and susceptibility to client drift. It enhances scalability, robustness, and efficiency in real-world settings with heterogeneous client populations and dynamic network conditions. Our results underscore the potential of AFL to drive advancements in distributed learning systems, particularly for large-scale, privacy-preserving applications in resource-constrained environments.

This paper addresses the critical problem of stabilizing an inherently unstable bicycle robot at low or zero speeds using advanced control strategies. The key objective is to develop a feedback controller that actively operates a handlebar, as well as a reaction wheel, maintaining balance while keeping performance indices low, including energy consumption. It is the energy consumption that makes the platform attractive for the users by extending the time of operation, not only due to the actions connected to motor design and battery design, but to the control system itself. To design and verify the control law that enables stabilization, and reduces energy consumption, bicycle dynamics and controllers based on linear quadratic regulator (LQR) techniques are designed, along with their robustified variants accounting for actuator failures and model uncertainties, and an innovative event-triggered control (ETC) scheme to reduce control updates and energy use. Simulations demonstrate effective stabilization from various initial lean angles under realistic conditions, with robustness notably improved at zero speed, where balancing is most challenging. Extending control to include both reaction wheel and handlebar actuation enhances dynamic maneuverability and energy efficiency. The event-triggered approach significantly lowers computational and actuator energy demands by updating controls only when needed, reducing mechanical wear. This research contributes to electromobility by improving the control and energy efficiency of personal transport devices like electric bicycles and autonomous bicycles, which rely on stable balancing mechanisms for safety and user comfort. The findings support future experimental validation and IoT implementation, pushing forward controlled, energy-conscious stabilization for emerging personal electromobility solutions.

Dynamic Programming suffers from the curse of dimensionality due to large state and action spaces, a challenge further compounded by uncertainties in the environment. To mitigate these issue, we explore an off-policy based Temporal Difference Approximate Dynamic Programming approach that preserves contraction mapping when projecting the problem into a subspace of selected features, accounting for the probability distribution of the perturbed transition probability matrix. We further demonstrate how this Approximate Dynamic Programming approach can be implemented as a particular variant of the Temporal Difference learning algorithm, adapted for handling perturbations. To validate our theoretical findings, we provide a numerical example using a Markov Decision Process corresponding to a resource allocation problem.