查看更多>>摘要：Edge Artificial Intelligence (AI) enables us to deploy distributed AI models, optimize computational and energy resources, minimize communication demands, and, most importantly, meet privacy requirements for Internet of Things (IoT) applications. Since data remains on the end-devices and only model parameters are shared with the server, it becomes possible to leverage the vast amount of data collected from smartphones and IoT devices without compromising the user's privacy. However, Federated Learning (FL) solutions also have well-known limitations. In particular, as systems that account for human behaviour become increasingly vital, future technologies need to become attuned to human behaviours. Indeed, we are already witnessing unparalleled advancements in technology that empower our tools and devices with intelligence, sensory abilities, and communication features. At the same time, continued advances in the miniaturization of computational capabilities can enable us to go far beyond the simple tagging and identification, towards integrating computational resources directly into these objects, thus making our tools “intelligent”. Yet, there is limited scientific work that considers humans as an integral part of these IoT-powered cyber-physical systems.
Cláudio G. S. CapanemaAllan M. de SouzaJoahannes B. D. da CostaFabrício A. Silva...
5-21页
查看更多>>摘要：Researchers have studied how to improve Federated Learning (FL) in various areas, such as statistical and system heterogeneity, communication cost, and privacy. So far, most of the proposed solutions are either very tied to the application context or complex to be broadly reproduced in real-life applications involving humans. Developing modular solutions that can be leveraged by the vast majority of FL structures and are independent of the application people use is the new research direction opened by this paper. In this work, we propose a plugin (named FedPredict) to address three problems simultaneously: data heterogeneity, low performance of new/untrained and/or outdated clients, and communication cost. We do so mainly by combining global and local parameters (which brings generalization and personalization) in the inference step while adapting layer selection and matrix factorization techniques to reduce the downlink communication cost (server to client). Due to its simplicity, it can be applied to federated learning of different number of topologies. Results show that adding the proposed plugin to a given FL solution can significantly reduce the downlink communication cost by up to 83.3% and improve accuracy by up to 304% compared to the original solution.
查看更多>>摘要：The proliferation of IoT devices, advancements in edge computing, and innovations in AI technology have created an ideal environment for the birth and growth of Edge AI. With the trend towards the Internet of Everything (IoE), the EdgeAI- Human-IoT architectural framework highlights the necessity for efficient data exchange interconnectivity. Ensuring secure data sharing and efficient data storage are pivotal challenges in achieving seamless data interconnection. Owing to its simplicity, ease of deployment, and consensus-reaching capabilities, the RAFT consensus algorithm, which is commonly used in distributed storage, faces limitations as the IoT scale expands. The computational, communication, and storage capabilities of nodes are constraints, and the security of data remains a concern. To address these complex challenges, we introduce the X-RAFT consensus algorithm, which is tailored for blockchain technology. This algorithm enhances system performance and robustness, mitigates the impact of system load, enhances system sustainability, and increases Byzantine fault tolerance. Through analysis and simulations, our proposed solution has been evidenced to provide reliable security and efficient performance.
Eric SturzingerJan HarkesPadmanabhan PillaiMahadev Satyanarayanan...
34-47页
查看更多>>摘要：We introduce survival-critical machine learning (SCML), in which a robot encounters dynamically evolving threats that it recognizes via machine learning (ML), and then neutralizes. We model survivability in SCML, and show the value of the recently developed approach of Live Learning. This edge-based ML technique embodies an iterative human-in-the-loop workflow that concurrently enlarges the training set, trains the next model in a sequence of “best-so-far” models, and performs inferencing for both threat detection and pseudo-labeling. We present experimental results using datasets from the domains of drone surveillance, planetary exploration, and underwater sensing to quantify the effectiveness of Live Learning as a mechanism for SCML.
Enrique Mármol CamposAurora Gonzalez-VidalJosé L. Hernández-RamosAntonio Skarmeta...
48-67页
查看更多>>摘要：Federated Learning (FL) represents a promising approach to typical privacy concerns associated with centralized Machine Learning (ML) deployments. Despite its well-known advantages, FL is vulnerable to security attacks such as Byzantine behaviors and poisoning attacks, which can significantly degrade model performance and hinder convergence. The effectiveness of existing approaches to mitigate complex attacks, such as median, trimmed mean, or Krum aggregation functions, has been only partially demonstrated in the case of specific attacks. Our study introduces a novel robust aggregation mechanism utilizing the Fourier Transform (FT), which is able to effectively handle sophisticated attacks without prior knowledge of the number of attackers. Employing this data technique, weights generated by FL clients are projected into the frequency domain to ascertain their density function, selecting the one exhibiting the highest frequency. Consequently, malicious clients’ weights are excluded. Our proposed approach was tested against various model poisoning attacks, demonstrating superior performance over state-of-the-art aggregation methods.
Quy Vu KhanhAbdellah ChehriVan Anh DangQuy Nguyen Minh...
68-79页
查看更多>>摘要：Across all periods of human history, the importance attributed to health has remained a fundamental and significant facet. This statement holds greater validity within the present context. The pressing demand for healthcare solutions with real-time capabilities, affordability, and high precision is crucial in medical research and technology progress. In recent times, there has been a significant advancement in emerging technologies such as AI, IoT, blockchain, and edge computing. These breakthrough developments have led to the creation of various intelligent applications. Smart healthcare applications can be realized by combining robust AI detection and prediction capabilities with edge computing architecture, which offers low computing costs and latency. In this paper, we begin by conducting a literature review of AI-assisted EC-based smart healthcare applications from the past three years. Our goal is to identify gaps and barriers in this field. We propose a smart healthcare architecture model that integrates AI technology into the edge. Finally, we summarize the challenges and research directions associated with the proposed model.
Tien NguyenAengus DalySergi Gomez-QuintanaFeargal O'Sullivan...
80-89页
查看更多>>摘要：Detecting seizures in neonates requires continuous electroencephalography (EEG) monitoring, a costly process that demands trained experts. Although recent advancements in machine learning offer promising solutions for automated seizure detection, the opaque nature of these algorithms poses significant challenges to their adoption in healthcare settings. A prior study demonstrated that integrating machine learning with sonification—an interpretation method that converts bio-signals into sound—can mitigate the black-box problem while enhancing seizure detection performance. This AI-assisted sonification algorithm can provide a valuable complementary tool in seizure monitoring besides the traditional visualization method. A low-power and affordable implementation of the algorithm is presented in this study using a microcontroller. To improve its practicality, we also introduce a real-time design that allows the sonification algorithm to function in parallel with data acquisition. The system consumes 12 mW in average, making it suitable for a battery-powered device.
NETL
NSTL
IEEE