Search

International audience; Recent medical applications are largely dominated by the application of Machine Learning (ML) models to assist expert decisions, leading to disruptive innovations in radiology, pathology, genomics, and hence modern healthcare systems in general. Despite the profitable usage of AI-based algorithms, these data-driven methods are facing issues such as the scarcity and privacy of user data, as well as the difficulty of institutions exchanging medical information. With insufficient data, ML is prevented from reaching its full potential, which is only possible if the database consists of the full spectrum of possible anatomies, pathologies, and input data types. To solve these issues, Federated Learning (FL) appeared as a valuable approach in the medical field, allowing patient data to stay where it is generated. Since an FL setting allows many clients to collaboratively train a model while keeping training data decentralized, it can protect privacy-sensitive medical data. However, FL is still unable to deliver all its promises and meets the more stringent requirements (e.g., latency, security) of a healthcare system based on multiple Internet of Medical Things (IoMT). For example, although no data are shared among the participants by definition in FL systems, some security risks are still present and can be considered as vulnerabilities from multiple aspects. This paper sheds light upon the emerging deployment of FL, provides a broad overview of current approaches and existing challenges, and outlines several directions of future work that are relevant to solving existing problems in federated healthcare, with a particular focus on security and privacy issues.

The abiding attempt of automation has also permeated the networks, with the ability to measure, analyze, and control themselves in an automated manner, by reacting to changes in the environment (e.g., demand). When provided with these features, networks are often labeled as “self-driving” or “autonomous”. In this regard, the provision and orchestration of physical or virtual resources are crucial for both Quality of Service (QoS) guarantees and cost management in the edge/cloud computing environment. To effectively manage the lifecycle of these resources, an auto-scaling mechanism is essential. However, traditional threshold-based and recent Machine Learning (ML)-based policies are often unable to address the soaring complexity of networks due to their centralized approach. By relying on multi-agent reinforcement learning, we propose Mystique, a solution that learns from the load on links to establish the minimal set of active network resources. As traffic demands ebb and flow, our adaptive and self-driving solution can scale up and down and also react to failures in a fully automated, flexible, and efficient manner. Our results demonstrate that the presented solution can reduce network energy consumption while providing an adequate service level, outperforming other benchmark auto-scaling approaches.

The recent growth of IoT devices, along with edge computing, has revealed many opportunities for novel applications. Among them, Unmanned Aerial Vehicles (UAVs), which are deployed for surveillance and environmental monitoring, are attracting increasing attention. In this context, typical solutions must deal with events that may change the state of the network, providing a service that continuously maintains a high level of performance. In this paper, we address this problem by proposing a distributed architecture that leverages a Multi-Agent Reinforcement Learning (MARL) technique to dynamically offload tasks from UAVs to the edge cloud. Nodes of the system co-operate to jointly minimize the overall latency perceived by the user and the energy usage on UAVs by continuously learning from the environment the best action, which entails the decision of offloading and, in this case, the best transmission technology, i.e., Wi-Fi or cellular. Results validate our distributed architecture and show the effectiveness of the approach in reaching the above targets.

Years of research on transport protocols have not solved the tussle between in-network and end-to-end congestion control. This debate is due to the variance of conditions and assumptions in different network scenarios, e.g., cellular versus data center networks. Recently, the community has proposed a few transport protocols driven by machine learning, nonetheless limited to end-to-end approaches.In this paper, we present Owl, a transport protocol based on reinforcement learning, whose goal is to select the proper congestion window learning from end-to-end features and network signals, when available. We show that our solution converges to a fair resource allocation after the learning overhead. Our kernel implementation, deployed over emulated and large scale virtual network testbeds, outperforms all benchmark solutions based on end-to-end or in-network congestion control.

In many edge computing applications, Unmanned Aerial Vehicles (UAVs) are required to be coordinated to perform several tasks. Each task is usually modeled as a process that a UAV runs, and could include hovering an area to find survivors after a natural disaster or sense and preprocess an image in cooperation with the edge cloud. Optimally and rapidly (re)assigning tasks to such IoT agents as the network conditions fluctuate and the battery of these agents quickly drains is a challenging problem. Existing solutions designed to proactively offload tasks are either energy unaware or they require solving computationally intensive task, and hence are less portable on constrained IoT devices. In this paper, we propose RITMO, a distributed and adaptive task offloading algorithm that aims to solve these challenges. RITMO exploits a simple yet effective regressor to dynamically predict the length of future UAV task queues. Such prediction is then used to anticipate the node overloading and avoid agents that are likely to exhaust their battery or their computational resources. Our results demonstrate how RITMO helps reduce the overall latency perceived by the application and the energy consumed by the nodes, outperforming recent solutions.

The edge computing paradigm allows computationally intensive tasks to be offloaded from small devices to nearby (more) powerful servers, via an edge network. The intersection between such edge computing paradigm and Machine Learning (ML), in general, and deep learning in particular, has brought to light several advantages for network operators: from automating management tasks, to gain additional insights on their networks. Most of the existing approaches that use ML to drive routing and traffic control decisions are valuable but rarely focus on challenged networks, that are characterized by continually varying network conditions and the high volume of traffic generated by edge devices. In particular, recently proposed distributed ML-based architectures require either a long synchronization phase or a training phase that is unsustainable for challenged networks. In this paper, we fill this knowledge gap with Blaster, a federated architecture for routing packets within a distributed edge network, to improve the application's performance and allow scalability of data-intensive applications. We also propose a novel path selection model that uses Long Short Term Memory (LSTM) to predict the optimal route. Finally, we present some initial results obtained by testing our approach via simulations and with a prototype deployed over the GENI testbed. By leveraging a Federated Learning (FL) model, our approach shows that we can optimize the communication between SDN controllers, preserving bandwidth for the data traffic.

Edge computing, combined with the proliferation of IoT devices, is generating new business model opportunities and applications. Among those applications, Unmanned Aerial Vehicles (UAVs) have been deployed in several scenarios, from surveillance and monitoring to disaster response, to precision agriculture. To support such applications, however, edge network managers and application programmers need to overcome a few challenges, e.g., unstable network conditions, high loss rate, and node failures. Existing solutions designed to mitigate such inefficiencies by predicting future network conditions are often computationally intensive and hence less portable on constrained devices. In this paper, we propose RITMO, a distributed and adaptive task planning algorithm that aims at solving these challenges while running on a network of UAV devices. We model our system as a network of queues, and we exploit a simple yet effective ARIMA regressor, to dynamically predict the length of future UAV task queues. Such prediction is then used to proactively migrate the tasks in case of a failure or unbalanced loads. Our simulation results demonstrate how RITMO helps to reduce the overall latency perceived by the application and anticipates the node overloading by avoiding agents that are likely to exhaust their computational resources.

The proliferation of the Internet of Things (IoT) in conjunction with edge computing has recently opened up several possibilities for several new applications. Typical examples are Unmanned Aerial Vehicles (UAV) that are deployed for rapid disaster response, photogrammetry, surveillance, and environmental monitoring. To support the flourishing development of Machine Learning assisted applications across all these networked applications, a common challenge is the provision of a persistent service, i.e., a service capable of consistently maintaining a high level of performance, facing possible failures. To address these service resilient challenges, we propose APRON, an edge solution for distributed and adaptive task planning management in a network of IoT devices, e.g., drones. Exploiting Jackson’s network model, our architecture applies a novel planning strategy to better support control and monitoring operations while the states of the network evolve. To demonstrate the functionalities of our architecture, we also implemented a deep-learning based audio-recognition application using the APRON NorthBound interface, to detect human voices in challenged networks. The application’s logic uses Transfer Learning to improve the audio classification accuracy and the runtime of the UAV-based rescue operations.

Flexible and transparent substrates are emerging as low cost and easy-to-operate support for surface-enhanced Raman spectroscopy (SERS). In particular, in situ SERS detection approach for surface characterization in transmission modality can be efficiently employed for non-invasive analysis of non-planar surfaces. Here we propose a new methodology to fabricate a homogenous, transparent, and flexible SERS membrane by the assistance of a thin TiO2 porous layer deposited on the PDMS surface, which supports the uniform loading of gold nanoparticles over large area. The substrate was first characterized for homogeneity, sensitivity and repeatability using a model molecule for SERS, i.e., 7-mercapto-4-methylcoumarin. Satisfactory intra-substrate uniformity and inter-substrates repeatability was achieved, showing an RSD of 10%, and an analytical sensitivity down to 10 nM was determined with an EF of 3.4 &times, 105 &plusmn, 0.4 &times, 105. Furthermore, SERS detection of pyrimethanil (PMT), a commonly employed pesticide in crops for human consumption, was performed in situ, exploiting the optical transparency of the device, using both model surfaces and non-flat bio-samples. PMT contamination at the phytochemical concentration levels corresponding to commonly used infield doses was successfully detected on the surface of the yellow Ficus benjiamina leaves, supporting the use of this substrate for food safety in-field application.

Recently, the growth of Internet of Things (IoT) devices combined with edge computing opened many opportunities for several novel applications. Typical examples are Unmanned Aerial Vehicles (UAV) that are deployed for photogrammetry, surveillance, disaster rapid response and environmental monitoring. A common challenge across all these networked applications is the ability to provide a persistent service - a service able to continuously maintain a high level of performance - responding to events that may change the state of the network, e.g., nodes or link failures. To cope with this challenge, in this paper we propose APRON, an edge cloud-assisted architecture for distributed and adaptive task planning management in a network of IoT devices, e.g., drones. APRON uses a novel planning strategy that, leveraging a Jackson's network model, supports monitoring and control operations while the states of the (edge or cloud) network evolve. By using APRON, edge computing application programmers can design and implement a wide range of IoT task management policies leveraging different protection methodologies across several failure models.

Tip-Enhanced Raman Spectroscopy (TERS) is a topographic and chemical analysis technique with nanoscale resolution, consisting of the combination of Scanning Probe Microscopy (SPM) and Localized Surface Plasmon Resonance (LSPR) for the enhancement of Raman scattering in the vicinity of the probe. The quantification of spatial resolution represents an important issue, and, as of now, standards for calibration are not available. In the present work a candidate reference sample for TERS measurements was fabricated. It consists of a flat, conductive gold surface with a nanometric grating of a self-assembled monolayer of Raman-active organic molecules fabricated by an optimized Electron Beam Lithography (EBL) method to replicate established SPM calibration standards. Its feasibility as a TERS standard was tested by STM-TERS imaging.