1,721,002 research outputs found
Stability and Wardrop Equilibria of Non-Cooperative Routing With Time-Varying Load
No full textNon-cooperative or selfish routing problems emerge in several applications of network control theory. Considering a multi-commodity setting subject to time-varying traffic demands, this paper studies the convergence properties of a family of non-cooperative routing control laws, originally developed in the literature for constant traffic demands. By employing results from hybrid systems theory and set stability, this paper identifies the minimum time between bounded load variations to assure the convergence of the controlled system towards a set of approximated Wardrop equilibria. Numerical simulations validate the results on a test scenario
Bellman's principle of optimality and deep reinforcement learning for time-varying tasks
Full text linkThis paper presents the first framework (up to the authors' knowledge) to address time-varying objectives in finite-horizon Deep Reinforcement Learning (DeepRL), based on a switching control solution developed on the ground of Bellman's principle of optimality. By augmenting the state space of the system with information on its visit time, the DeepRL agent is able to solve problems in which its task dynamically changes within the same episode. To address the scalability problems caused by the state space augmentation, we propose a procedure to partition the episode length to define separate sub-problems that are then solved by specialised DeepRL agents. Contrary to standard solutions, with the proposed approach the DeepRL agents correctly estimate the value function at each time-step and are hence able to solve time-varying tasks. Numerical simulations validate the approach in a classic RL environment
Smart Healthy Schools: An IoT-enabled concept for multi-room dynamic air quality control
Full text linkSmart Healthy Schools (SHS) are a new paradigm in building engineering and infection risk control in school buildings where the disciplines of Indoor Air Quality (IAQ), IoT (Internet of Things) and Artificial Intelligence (AI) merge together. In the post-pandemic era, equipping schools with a network of smart IoT sensors has become critical to aspire for the optimal control of the IAQ and lowering the airborne infection risk of several pathogens, indirectly related to cumulated human emitted CO2 levels over time. Thermal energy waste in winter due to improved air renewal remains of major concern but can be well monitored within a SHS monitoring architecture thanks to the flexibility of the LoRaWAN protocol able to process also a large amount of energy and climatic data at room and building scale. In this work, we report the design of the AulaSicura platform, an IoT control system co-designed by the main author and Gizero Energie to implement the SHS paradigm via clearly visible (and audible) alarm signalling in existing and new school buildings. The cloud-based LoRa system is capable of continuous and simultaneous monitoring of a variety of sensors and IAQ parameters including indoor/oudoor temperatures, rel. humidities and human-emitted excess CO2. The multi-room monitoring concept of indoor-CO2 levels allows centralized control of natural ventilation levels in individual classrooms and can handle (quasi)-real-time data, relevant for data post-processing and future developments in (quasi)-real-rime assessment of IAQ and infection risk levels at single room scale. The sensor network is also extensible to up to one thousand of classrooms per LoRa-node allowing centralized control of entire school districts at an urban scale. Moreover, through Modbus-LoRa I/O converters, AulaSicura can also control the same amount of mechanical ventilation units per node either in pure or hybrid mechanical ventilation modes
Robust and fault-tolerant spacecraft attitude control based on an extended-observer design
Full text linkThe aim of this work is to develop a robust control strategy able to drive the attitude of a spacecraft to a reference value, despite the presence of unknown but bounded uncertainties in the system parameters and external disturbances. Thanks to the use of an extended observer design, the proposed control law is robust against all the uncertainties that affect the high-frequency gain matrix, which is shown to capture a broad spectrum of modelling issues, some of which are often neglected by traditional approaches. The proposed controller then provides robustness against parametric uncertainties, as moment of inertia estimation, payload deformations, actuator faults and external disturbances, while maintaining its asymptotic properties
A Weighted Average Consensus Approach for Decentralized Federated Learning
Full text linkFederated learning (FedL) is a machine learning (ML) technique utilized to train deep neural networks (DeepNNs) in a distributed way without the need to share data among the federated training clients. FedL was proposed for edge computing and Internet of things (IoT) tasks in which a centralized server was responsible for coordinating and governing the training process. To remove the
design limitation implied by the centralized entity, this work proposes two different solutions to decentralize existing FedL algorithms, enabling the application of FedL on networks with arbitrary communication topologies, and thus extending the domain of application of FedL to more complex scenarios and new tasks. Of the two proposed algorithms, one, called FedLCon, is developed based on results from discrete-time weighted average consensus theory and is able to reconstruct the performances of the standard centralized FedL solutions, as also shown by the reported validation tests
Automated Optical Inspection for Printed Circuit Board Assembly Manufacturing with Transfer Learning and Synthetic Data Generation
Full text linkAutomated Optical Inspection (AOI) is among the most common and effective quality checks employed in production lines. This paper details the design of a Deep Learning solution that was developed for addressing a specific quality control in a Printed Circuit Board Assembly (PCBA) manufacturing process. The developed Deep Neural Network exploits transfer learning and a synthetic data generation process to be trained even if the quantity of the data samples available is low. The overall AOI system was designed to be deployed on low-cost hardware with limited computing capabilities to ease its deployment in industrial settings
Stability of Non-Cooperative Load Balancing with Time-Varying Latency
Full text linkThe problem of non-cooperative load balancing arises in multi-agent scenarios where users/services compete for some limited resources. This study, leveraging on results from set stability and switched systems control theory, analyses the convergence properties of a class of load-balancing strategies towards a set of approximated non-cooperative equilibria in a scenario in which the performance of the resource providers is described by a time-varying latency function
Chance-Constrained Control with Lexicographic Deep Reinforcement Learning
Full text linkThis paper proposes a lexicographic Deep Reinforcement Learning (DeepRL)-based approach to chance-constrained Markov Decision Processes, in which the controller seeks to ensure that the probability of satisfying the constraint is above a given threshold. Standard DeepRL approaches require i) the constraints to be included as additional weighted terms in the cost function, in a multi-objective fashion, and ii) the tuning of the introduced weights during the training phase of the Deep Neural Network (DNN) according to the probability thresholds. The proposed approach, instead, requires to separately train one constraint-free DNN and one DNN associated to each constraint and then, at each time-step, to select which DNN to use depending on the system observed state. The presented solution does not require any hyper-parameter tuning besides the standard DNN ones, even if the probability thresholds changes. A lexicographic version of the well-known DeepRL algorithm DQN is also proposed and validated via simulations
Deep Image Inpainting to Support Endoscopic Procedures
No full textDeep image inpainting is a computer vision task that uses Deep Neural Networks to generate plausible content to complete an image, for example for the restoration of a damaged image or the removal of unwanted elements captured in the picture. This paper uses deep image inpainting to restore endoscopic images that are affected by various types of artifacts. To this end, we developed a transfer learning-based procedure that uses the CSA inpainting model, which was originally proposed for unrelated tasks including the restoration of images from the Paris StreetView Dataset. The proposed system is trained and validated on the EndoCV2020 dataset, consisting of images from real endoscopies, highlighting how deep image inpainting may be a promising technology for frame restoration during medical procedures
A Discrete-Time Multi-Hop Consensus Protocol for Decentralized Federated Learning
Full text linkThis paper presents a Federated Learning (FL) algorithm that allows the decentralization of all FL solutions that employ a model-averaging procedure. The proposed algorithm proves to be capable of attaining faster convergence rates and no performance loss against the starting centralized FL implementation with a reduced communication overhead compared to existing consensus-based and centralized solutions. To this end, a Multi-Hop consensus protocol, originally presented in the scope of dynamical system consensus theory, leveraging on standard Lyapunov stability discussions, has been proposed to assure that all federation clients share the same average model employing only information obtained from their m-step neighbours. Experimental results on different communication topologies and the MNIST and MedMNIST v2 datasets validate the algorithm properties demonstrating a performance drop, compared with centralized FL setting, of about 1%
- …
