Federated Learning (FL) aims to train artificial intelligence models without the need to share private raw data, thereby preserving privacy and security. Typically, it is assumed that all participating FL clients will act honestly to develop an accurate model. However, some clients may behave deceptively, manipulating their data to bias the model’s predictions and also degrade its generalization ability. This paper addresses the issue of fairness in FL from the perspective of client truthfulness. We introduce Federated Objective (FedObj), a novel aggregation method designed to minimize the impact of malicious clients and thereby improve the overall model’s robustness to such behavior. Our results show that FedObj achieves state-of-the-art performance in standard scenarios and outperforms conventional strategies when deceptive clients are involved. FedObj is a valuable approach for the collaborative development of trustworthy and fair AI systems, as it is significantly resilient to the misleading practices of malicious FL clients.

Federated Learning exploits local model training to aggregate and create a global model without sharing raw data. Each client trains a local model and shares it to aggregate a global one. Several works demonstrate that starting from trained weights, it’s possible to reconstruct the original used data. For this reason, the research and industrial world introduced the Homomorphic Encryption technique to encrypt the transmitted local model’s weights. This approach protects the trained weights from a hypothetical malicious aggregator server that can not perform operations over plaintext weights. In the proposed work, we implement a Federated Learning solution applying Homomorphic Encryption using the Flower framework and the Tenseal library. Our solution follows best practices for custom aggregation strategies with the Flower framework, making it possible to provide it to the community.

Federated Learning lays its foundation on the computation and availability of data provided by the increasingly popular Edge devices. By performing on-device training, privacy over the data can be ensured and the computational capacity of the devices is directly exploited. On the other hand, the heterogeneous characteristics of such devices and the difficulty of establishing a proper communication network are often complex obstacles to overcome when implementing Federated Learning solutions. Leveraging ONNX Training Web, WebAssembly, and the Cloud-Edge-Client Continuum concept, we propose FLAT, a system that allows Federated Learning algorithms to run seamlessly on Web browsers, in the form of microservices, without the need for any dependencies and configuration. Through this approach, any device equipped with a browser, even in headless mode, can dynamically join a federated learning cluster in plug-and-play mode by simply connecting to a Web address. The effectiveness of the system is tested with the MNIST and CIFAR10 datasets, using respectively a DNN and a CNN with FedAvg as the aggregation strategy and considering different devices and browsers.

Internet of Things (IoT) became more and more popular because of the raise of ubiquitous internet connected devices. In this regard, IoT nodes are often organized in wireless sensor networks to facilitate communication and perform a coral computation. Such a network is often employed in urban or rural areas, i.e., for traffic, fires, and floods monitoring. Nodes are, therefore, deployed in remote areas, preventing the possibility to frequently access them, i.e., for firmware update. In this context, over-the-air (OTA) firmware update is used to remotely change the behavior of one or more nodes. In this paper, we firstly build a wireless mesh network with microcontrollers (i.e., ESP32) and, therefore, propose a secure filtered O TA firmware update involving firmware freshness (i.e., quarantine when firmware is not up-to-date), key pairing, and digital signature for data integrity and non-repudiation. The system is evaluated in terms of deactivation time $(s)$, energy consumption $(kWh)$, and greenhouse gases $(\mathrm{kgCO}_{2}\mathrm{e})$, highlighting good results in terms of scalability for grouped updates.

Digital Twins (DTs) in Metaverse face many challenges such as the lack of optimized AI models to allow the interaction between the user and the virtual environment. In this paper, we propose an optimized model for human language processing based on Convolutional Neural Networks (CCNs) and we present an input processing strategy to meet the real-time requirements of smart applications that integrate DTs oriented to speech-based functionalities for user interaction and Metaverse. In our solution, CNNs are applied for the processing and classification of the human voice, while structured data and MFCC coefficients are used to train the neural networks and generate interference in the models. Similarly, the MFCC algorithm is provided to extract the unique characteristics that specify each generated audio file and to reduce the complexity of the neural network model in order to obtain better performance. Starting from an approach to the problem available in the literature, we have optimized a specific CNN model for Natural Language Processing (NLP) in order to increase effective results. The proposed model has demonstrated excellent performance and can be used as a basis for the implementation of software that allows the interaction of DTs with voice commands issued by a user.

Edge systems are increasingly popular for data collection and processing. Typically, due to their limited storage capacity, pieces of data are continuously exchanged with Cloud systems which store them in distributed DataBase Management System (DBMS). This scenario, known as Cloud/Edge Continuum, is critical from a data security point of view as it is exposed to many risks. Transparent Data Encryption (TDE) is proposed as a possible solution for encrypting database files. However, current solutions do not suit the Cloud/Edge continuum requirements. In this paper, we aim at fulfilling this gap by proposing a solution to encrypt the data locally at the Edge and transfer them to a distributed database over the Cloud. Our approach allows us to perform queries directly on encrypted data over the Cloud and to retrieve them on the Edge for decryption. Experiments performed on different NoSQL DBMS solutions demonstrate the feasibility of our approach.

TEMA project is a Horizon Europe funded project that aims at addressing Natural Disaster Management by the use of sophisticated Cloud-Edge Continuum infrastructures by means of data analysis algorithms wrapped in Serverless functions deployed on a distributed infrastructure according to a Federated Learning scheduler that constantly monitors the infrastructure in search of the best way to satisfy required QoS constraints. In this paper, we discuss the advantages of Serverless workflow and how they can be used and monitored to natively trigger complex algorithm pipelines in the continuum, dynamically placing and relocating them taking into account incoming IoT data, QoS constraints, and the current status of the continuum infrastructure. Therefore we presented the Urgent Function Enabler (UFE) platform, a fully distributed architecture able to define, spread, and manage FaaS functions, using local IOT data managed using the Fiware ecosystem and a computing infrastructure composed of mobile and stable nodes.

Nowadays Osmotic Computing is emerging as one of the paradigms used to guarantee the Cloud Continuum, and this popularity is strictly related to the capacity to embrace inside it some hot topics like containers, microservices, orchestration and Function as a Service (FaaS). The Osmotic principle is quite simple, it aims to create a federated heterogeneous infrastructure, where an application's components can smoothly move following a concentration rule. In this work, we aim to solve two big constraints of Osmotic Computing related to the incapacity to manage dynamic access rules for accessing the applications inside the Osmotic Infrastructure and the incapacity to keep alive and secure the access to these applications even in presence of network disconnections. For overcoming these limits we designed and implemented a new Osmotic component, that acts as an eventually consistent distributed peer to peer access management system. This new component is used to keep a local Identity and Access Manager (IAM) that permits at any time to access the resource available in an Osmotic node and to update the access rules that allow or deny access to hosted applications. This component has been already integrated inside a Kubernetes based Osmotic Infrastructure and we presented two typical use cases where it can be exploited.

Recently, both scientific and industrial communities have highlighted the importance to run Machine Learning (ML) applications on Edge computing closer to the end-user and to managed raw data, for many reasons including quality of service (QoS) and security. However, due to the limited computing, storage and network resources at the Edge, several ML algorithms have been re-designed to be deployed on Edge devices. In this paper, we want to explore in detail Edge Federation for supporting ML-based solutions. In particular, we present a new platform for the deployment and the management of complex services at the Edge. It provides an interface that allows us to arrange applications as a collection of interconnected lightweight loosely-coupled services (i.e., microservices) and enables their management across Federated Edge devices through the abstraction of the underlying clusters of physical devices. The proposed solution is validated by a use case related to video analysis in the morphological field.

Sustainable mobility is a key objective for many Smart Cities. In this paper, we present an application for sustainable mobility that aims at encouraging citizens to use low-impact vehicles instead of private cars. Through a partnership between the University Messina and the Municipality of the Messina city (Italy), we developed MuoviME, a digital application to assign citizens electric bikes, free of charge for a limited period of time. The key issue we addressed in the development of such an application is security, both in terms of secure authentication of citizens that access the service and tracking of the whole assignment process, from the user's bicycle request to its restitution. To achieve this goal, we implemented a solution for the physical recognition of the user based on two-factor authentication (2FA) and Blockchain technology. This paper summarizes the secure by design approach, implementation details, and some experimental results on the service efficiency.

Nowadays, the increasing complexity of digital applications for social and business activities has required more and more advanced mechanisms to prove the identity of subjects like those based on the Two-Factor Authentication (2FA). Such an approach improves the typical authentication paradigm but it has still some weaknesses. Specifically, it has to deal with the disadvantages of a centralized architecture causing several security threats like denial of service (DoS) and man-in-the-middle (MITM). In fact, an attacker who succeeds in violating the central authentication server could be able to impersonate an authorized user or block the whole service. This work advances the state of art of 2FA solutions by proposing a decentralized Microservices and Blockchain Based One Time Password (MBB-OTP) protocol for security-enhanced authentication able to mitigate the aforementioned threats and to fit different application scenarios. Experiments prove the goodness of our MBB-OTP protocol considering both private and public Blockchain configurations.