Human Activity Recognition (HAR) stands as a crucial technology, with applications ranging from healthcare monitoring to sports analytics. However, the traditional approach to HAR is often time-consuming and susceptible to human errors due to the high complexities involved in processing diverse sensor data. Recognizing the imperative for efficiency and accuracy in HAR systems, we propose the development of an Automated Decision-maker (ADM) system. This system serves to automate HAR pipelines, addressing the challenges posed by the huge sensor data. By harnessing the power of automation, ADM significantly streamlines the HAR process, reducing the time required for hyperparameter tuning and minimizing the risk of human errors. The results obtained from our proposed ADM system demonstrate notable improvements in HAR performance, showcasing achieved accuracy of 96.436% for UCI-HAR & 99.783% for PAMAP2 datasets. Moreover, ADM can be described as an innovative approach that contributes to the optimization of HAR systems while also establishing a foundation for building robust and reliable systems in complex environments.

The global maritime industry is undergoing a digital transformation driven by the Fourth Industrial Revolution, which gives rise to “smart ports” or “Port 4.0” environments. These ports leverage smart technologies, and thus increase exposure to cyber threats that have seen dramatic growth in recent years. This paper presents four key contributions to strengthen cybersecurity in smart Port ecosystems. First, it maps the current cyber-threat landscape across both Information Technology (IT) and Operational Technology (OT) systems used in the smart Port environment, highlighting critical vulnerabilities. Then, it analyzes existing regulatory and standard frameworks such as the International Maritime Organization (IMO) guidelines, ISO/IEC 27001, and the National Institute of Standards and Technology (NIST) cybersecurity framework, identifying alignment gaps with maritime operational realities. Further, this paper also provides structured, Port-specific cybersecurity recommendations tailored to the complex interplay of legacy OT systems and modern digital technologies. Finally, the paper discusses AI-assisted cybersecurity solutions available in the literature, highlighting how advanced AI-based analytics, predictive modeling, and automated incident response can be incorporated. The insights presented are intended to help Port authorities build resilient, adaptive cybersecurity postures in an increasingly interconnected maritime domain.

The Internet of Medical Things (IoMT) has revolutionized healthcare by enabling real-time patient monitoring, remote diagnostics, and intelligent decision-making. However, IoMT data is prone to unauthorized access, resulting in a significant loss of privacy and security. To address these challenges, we propose a novel security framework which is based on XChaCha20-Encryption fortified with a Role-Based Access Control (RBAC) mechanism and a blockchain-integrated ML model to establish a highly secure and intelligent healthcare ecosystem. The proposed framework employs a Proof of Authority (PoA) consensus mechanism to validate and secure blockchain operations, making it particularly well-suited for real time IoMT applications. The performance of the system is evaluated on the WUSTL-EHMS-2020 dataset, demonstrating superior results over other state-of-the-art approaches. Moreover, the proposed framework achieves remarkable encryption and decryption times of 0.61 and 0.71 seconds, respectively. Along with data privacy proposed framework outperforms other methods and achieves an accuracy of 99.43% for detecting cyber attacks launched against IoMT data.

Wastewater treatment plants (WWTPs) are among the most energy-intensive components of urban infrastructure and bear strict regulatory responsibilities for wastewater quality. These dual challenges, minimizing energy consumption and maintaining environmental compliance, are deeply interrelated and must be managed simultaneously to achieve sustainable plant operation. This study proposes a framework that comprises two customized components. The first component employs a voting ensemble model based on transformer architecture to predict energy consumption. It processes heterogeneous feature domains — including hydraulic, wastewater, and climatic variables — through parallel attention-driven streams. The outputs from these streams are then aggregated using a weighted voting mechanism to produce the final prediction. Second, a multitask Bidirectional Gated Recurrent Unit (Bi-GRU) forecasts wastewater quality indicators concurrently (ammonia, Biochemical Oxygen Demand (BOD), and Chemical Oxygen Demand (COD)), capturing shared temporal dependencies and reducing model complexity. A hybrid preprocessing strategy is applied, incorporating domain-aware outlier detection (z-score and Interquartile Range (IQR)), K-Nearest Neighbors (KNN) Imputation, and feature selection using Extreme Gradient Boosting (XGBoost). Experimental results showed that. The voting ensemble model achieved the best results for energy consumption prediction with 31.61 of Root Mean Squared Error (RMSE). The multitask Bi-GRU achieved the best results for wastewater quality indicators with RMSE at 6.1689, 48.0323, and 88.2214 for ammonia, BOD, and COD, respectively. This work is among the first to integrate transformer ensembles and multitask learning in a unified WWTP forecasting system. Simultaneously addressing energy efficiency and water quality assurance, this offers a practical, scalable, and intelligent decision-support tool for sustainable wastewater management.

ABSTRACT The application of artificial intelligence (AI) in healthcare has been witnessing an increasing interest. Particularly, federated learning (FL) has become favourable due to its potential for enhancing model quality whilst maintaining data privacy and security. However, the effectiveness of present FL methodologies could underperform under non‐IID conditions, characterised by divergent data distributions across clients. The globally constructed FL model may suffer potent issues by allowing the least‐performing models to equal participation. Thus, we propose a new accuracy‐based FL approach (FedAcc) which only takes into account the clients' validation accuracy to consider their participation during global aggregation, also called Smart Healthcare Amplified (SHA). However, with limited supervised data it is challenging to increase the model performance thus concept of transfer learning (TL) is used. TL enables the global model to integrate knowledge from precomputed systems, resulting in an efficient model. However, the complexity of the global system is amplified by these TL models, leading to challenges related to vanishing gradients, particularly when dealing with a substantial number of layers. To mitigate this, we present a Transfer Learning Domain Adaptation Model (TLDAM). TLDAM employs a two‐layered sequentially trained TL model, which contains approximately 50% fewer layers compared to traditional TL models. TLDAM is trained on multiple datasets such as MNIST and CIFAR10, to enhance its knowledge and make it domain‐adaptive. Moreover, experimental results conducted on the UCI‐HAR dataset reveal the supremacy of our proposed framework with an accuracy of 94.2990%, F‐score of 94.2820%, precision of 94.3058%, and recall of 94.2993% over traditional FL techniques and state‐of‐the‐art techniques.

Abstract The convergence of the Internet of Things (IoT) and Industrial Internet of Things (IIoT) within the Industry 4.0 paradigm leverages software-defined networking, multi-cloud architectures, and edge/fog computing to enhance industrial processes. However, this digital transformation introduces significant cybersecurity and privacy vulnerabilities within the complex, data-intensive IoT/IIoT ecosystems. To mitigate these risks, this research proposes a novel Anomaly-based Intrusion Detection System using Voting-based Ensemble Model (ABIDS-VEM) in Industry 4.0 environments. The VEM architecture synergistically combines multiple machine learning algorithms and gradient boosting frameworks, including CatBoost (CB), XGBoost (XGB), LightGBM (LGBM), Logistic Regression (LR), and Random Forest (RF), to enhance the precision and computational efficiency of intrusion detection systems (IDS) in IoT/IIoT contexts. The proposed framework incorporates a data ramification process, in which the data is divided into multiple parts, feature selection process which is optimized through the Equilibrium Optimizer (EO) algorithm, and outlier detection utilizing the Isolation Forest (IF) method. Comprehensive empirical evaluations were conducted using three benchmark datasets: XIIoTID, NSL-KDD, and UNSW-NB15, to validate the efficacy of the proposed system. The model achieves high accuracy across datasets: 98.1476% for XIIoT-ID, an impressive accuracy of 98.9671% for NSL-KDD, and 94.1327% for UNSW-NB15 dataset. These experimental results demonstrate the potential of this approach to significantly enhance the resilience of critical industrial systems and data against evolving cyber threats, thereby supporting the continued evolution of Industry 4.0 technologies and bolstering the security posture of IoT/IIoT ecosystems. This research contributes to the ongoing efforts to secure the rapidly expanding digital industrial landscape, offering a robust solution for detecting and mitigating sophisticated cyberattacks in the increasingly interconnected and data-driven industrial environments of the future.

The rapid integration of digital technologies has precipitated a substantial transition from traditional in-person commerce to digital trade, even encompassing physical goods. This transformation not only simplifies the purchasing process but also broadens the market reach for vendors, enabling global connectivity with consumers. Presently, there is a shift towards Metaverse-based applications, with blockchains serving as the foundational infrastructure for data management, thereby enhancing the overall user experience. Given this context, integrating payment methods becomes pivotal, particularly in light of the widespread adoption of blockchain technology. Our proposal introduces xPayment, a decentralised payment protocol designed to facilitate end-to-end and cross-domain transactions for digital goods via payment channel networks (PCNs). Initially, we introduce the concept of cyclic exchange and formally demonstrate its atomicity, on which we develop xPayment protocol to facilitate an independent and interoperable payment mechanism integrated into blockchain-based ownership transfers. We provide both theoretical and empirical evidence of its atomicity, minimal merchant opportunity cost, and customisable privacy features.

Cryptocurrency, despite the upsurge as a speculative investment, is still a long way from being people’s money. The extreme technicality poses a significant barrier for the general public to adopt it as a medium of exchange. Therefore, simplifying the payment process is a predominant necessity; for example, facilitation in adopting conventional payment instruments can bring convenience to both merchants and consumers. This article proposes a novel cryptocurrency card payment protocol leveraging the Payment Channel Network (PCN) concept, facilitating high throughput and affordable transactions. Our design is based on the Bitcoin-backed Lightning Network (LN) with adjustments to make it executable by a smart card. It also preserves the decentralized nature of cryptocurrencies, reduces operational costs in several ways, and allows instant settlement for the recipients as compared to both conventional and cryptocurrency card payment systems. Our game theoretical analysis, where we model the engagement between the cardholder and the card agent as a long-run extensive form game, attests that they accord with the protocol without experiencing any honest loss under pragmatic conditions. We also discuss the privacy features compared to conventional card payments and LN. The protocol can function independently, without the need for trust, and can also be regulated for those seeking government mediation. This approach can potentially revolutionize the payment landscape by allowing the public to use cryptocurrency for everyday transactions conveniently and allowing existing cryptocurrency holders to conduct affordable micropayments.

With the advent of Industry 4.0 (I4.0) leading to the proliferation of industrial process data, deep learning (DL) techniques have become instrumental in developing intelligent fault diagnosis (FD) applications. However, despite their potentially superior process monitoring capabilities, DL-based FD models are poorly calibrated and generate point estimate predictions without the associated uncertainty estimates. For DL-based FD models, accurate predictive uncertainty estimates from well-calibrated models are essential in ensuring industrial process safety and reliability. This article proposes ensemble-to-distribution (E2D), an uncertainty-aware combination method for quality monitoring FD based on an ensemble of deep neural networks. First, E2D addresses safety by providing accurate uncertainty estimates on model predictions, enabling informed decision-making to minimize operational risks. Second, E2D improves model performance on out-of-distribution detection tasks to facilitate deployments in the real world. Third, E2D is a post hoc application, implementable at inference time, and compatible with diverse pretrained models. Finally, to demonstrate the effectiveness of E2D, we explore the problem of monitoring the stability of industrial processes and product quality using case studies on the steel plates faults and APS failure at Scania trucks datasets.

The increasing importance of audio-based healthcare diagnostics, particularly in chronic respiratory problems, has prompted the development of novel approaches for disease classification using the full spectrum analysis of cough sounds. This paper presents an innovative study exploring the potential of employing supervised and semi-supervised learning methodologies for disease categorization based on cough audio samples. Specifically, this study focuses on scenarios characterized by a shortage of annotated data related to chronic diseases. The objectives of our study involved the utilization of standard machine learning algorithms for direct classification based on embeddings, as well as the integration of Graph Neural Networks (GNNs) on the KNN graph. Preliminary results indicate that GNN models consistently outperformed traditional classifiers. For instance, with just 1 % of the data, GAT and GCN achieved AUC PR values of 0.84 and 0.87, respectively, surpassing all traditional methods. The superiority mentioned above was sustained even when the data fraction was augmented to 3% and 5%. The usefulness of graph neural networks (GNNs) was further supported by comprehensive performance measures, wherein the graph convolutional network (GCN) exhibited exceptional preci-sion and PR (precision-recall). In summary, the AudioVecDiagnosis framework presents a promising opportunity for further investigation in audio-based healthcare diagnostics. It provides an optimum approach for situations with a limited availability of labeled data.

Product failure detection in smart manufacturing is important because it allows manufacturers to quickly identify and isolate faulty products before they reach the end of the production line. In today’s fast-paced and highly competitive business environment, manufacturers need to quickly and accurately identify product failures to stay competitive and meet customer expectations. Previously the product quality was inspected manually and now it is examined using a machine learning algorithm to overcome the limitations of manual inspection. However, in the latter case AI and ML experts are required to do the task. Therefore, to overcome such dependency, this work proposes a microservice to allow the end user (an industry person, as well as an automated hardware/software agent) to test different combinations of data science and AI tools and technologies without any AI expertise. The proposed microservice exposes APIs that make it possible to select different combinations of feature selection, sampling, and classification algorithm. A demo environment with a Postman collection that includes few API calls to demonstrate how the proposed module enables customized selection to detect faulty products, is also discussed.

Reducing product failure rates is crucial to ensure a healthy production line. However, the current approach for inspecting product quality is inefficient, costly, and time-consuming, relying on manual inspection at the end of the production process. This research paper focuses on the utilization of transfer learning, an intelligent machine-learning technique, to improve the accuracy and efficiency of product quality inspection in production lines. The proposed approach utilizes transfer learning to adapt a pre-trained model from a related domain to the target domain, enabling accurate product quality prediction with limited data. The reference architecture provides a framework for implementing the proposed approach in a manufacturing environment, enabling real-time monitoring and decision-making based on product quality predictions. The proposed approach can improve the accuracy of faulty product detection by up to 11% compared to traditional techniques, as demonstrated by evaluations on a real-world production dataset.

In modern industry, adaptation to market changes, as well as prompt reaction to a variety of predictable and unpredictable events, is a key requirement. Ubiquitous computing, real-time analytics, reconfigurable hardware/software components, often coexist in the complex, internally variegated, and often proprietary systems that are traditionally deployed to meet such requirement. However, such tailor-made systems meet only in part the requirements of openness, security, monitorability, geographical distribution, and most of all, remote extendability and changeability, which are crucial for prompt reaction to unforeseen circumstances. In this work, a containerized service application named Network Factory is presented. It enables the remote construction, configuration and operation of resilient computation systems that meet the above-mentioned requirements, and distinguish for their logical simplicity and for the uniform addressing of elaborations and human-computer interfaces, which are achieved through few reconfigurable components and communication mechanisms that are used from the production line up to the Cloud. Source code, documentation, and step-by-step introductory guides are publicly available in a dedicated GitHub repository, and distributed under the CC-BY-4.0 license.

Modern manufacturing systems characterize for the multiple dimensions of their complexity. They are numerically complex, as they consist of several components. They are logically complex, as multiple and variegated links exist among the different components. They are technologically complex, as a mix of different hardware and software technologies and architectures is typically found. They are geographically complex, as they often extend across multiple physical locations and sometimes involve multiple organizations. However, resilience to predictable and unpredictable events through timely, efficient, and effective reconfiguration of the whole manufacturing ecosystem remains a key objective, being it a key enabler of industry competitiveness. In this work, an innovative approach based on API request collections, containerization technologies, and past research about remotely reconfigurable distributed systems, is proposed for achieving ultimate resilience in modern industry.

Conventional Programmable Logic Controller (PLC) systems are becoming increasingly challenging to manage due to hardware and software dependencies. Moreover, the number and size of conventional PLCs on factory floors continue to increase, and virtualized PLC (vPLC) offers a solution to address these challenges. The utilization of vPLC offers the advantages of streamlining communication between high-level applications and low-level machine operations, enhancing programming ability in process control systems by abstracting control functions from I/O modules, and increasing automation in industrial control networks. Nevertheless, the connection of vPLC to the internet and cloud services presents a considerable cybersecurity risk, and the crucial aspect of information security for vPLCs is ensuring their availability. Distributed Denial of Service (DDoS) attacks can be particularly devastating for vPLCs, as they rely on internet connectivity to function. DDoS attacks on vPLC overwhelm it and causing it to become unavailable. vPLCs manages control systems and if targeted by a DDoS attack, these systems could become unresponsive, leading to significant disruption to industrial processes. Thus, implementing effective DDoS protection measures is crucial for ensuring the availability and reliability of vPLCs in industrial settings. Therefore, this work proposes a Federated learning enabled Threat Intelligence Unit (FedTIU) for detecting DDoS attacks on vPLCs on an Edge Compute Stack near to vPLC. The proposed approach involves collaborative model training using federated learning techniques to gain knowledge of new attack patterns from other industrial sites while maintaining data privacy.

Abstract The development of the Internet of Things (IoT) technology and their integration in smart cities have changed the way we work and live, and enriched our society. However, IoT technologies present several challenges such as increases in energy consumption, and produces toxic pollution as well as E-waste in smart cities. Smart city applications must be environmentally-friendly, hence require a move towards green IoT. Green IoT leads to an eco-friendly environment, which is more sustainable for smart cities. Therefore, it is essential to address the techniques and strategies for reducing pollution hazards, traffic waste, resource usage, energy consumption, providing public safety, life quality, and sustaining the environment and cost management. This survey focuses on providing a comprehensive review of the techniques and strategies for making cities smarter, sustainable, and eco-friendly. Furthermore, the survey focuses on IoT and its capabilities to merge into aspects of potential to address the needs of smart cities. Finally, we discuss challenges and opportunities for future research in smart city applications.

The electric vehicles (EVs) charging stations (CSs) at public premises have higher installation and power consumption costs. The potential benefits of public CSs rely on their efficient utilization. However, the conventional charging methods obligate a long waiting time and thereby deteriorate their efficiency with low utilization. This paper suggests a novel fuzzy integer linear programming and a heuristic fuzzy inference approach (FIA) for CSs utilization. The model introduces the underlying fuzzy inference system and a detailed formulation for obtaining the optimal solution. The developed fuzzy inference incorporates the uncertain and independent available power, required state-of-charge, and dwell time from the power grid and EVs domains and correlates them into weighted control variables. The FIA automates the service provision for the EVs with the most urgent requirements by resolving the objective function utilizing the weighted control variables, thereby optimizing the waiting time and the CSs utilization. To evaluate the effectiveness of the proposed FIA, several case studies were conducted, corresponding to different parking capacities and installations of CSs. Moreover, the simulations were conducted on EVs with varying battery capacities, and their performance was evaluated based on several metrics, including average waiting time, utilization of CSs, fairness, and execution time. The simulation results have confirmed that the effectiveness of the proposed FIA scheduling method is considerably higher than that of the other methods discussed.

Scene graph generation aims to capture the semantic elements in images by modelling objects and their relationships in a structured manner, which are essential for visual understanding and reasoning tasks including image captioning, visual question answering, multimedia event processing, visual storytelling and image retrieval. The existing scene graph generation approaches provide limited performance and expressiveness for higher-level visual understanding and reasoning. This challenge can be mitigated by leveraging commonsense knowledge, such as related facts and background knowledge, about the semantic elements in scene graphs. In this paper, we propose the infusion of diverse commonsense knowledge about the semantic elements in scene graphs to generate rich and expressive scene graphs using a heterogeneous knowledge source that contains commonsense knowledge consolidated from seven different knowledge bases. The graph embeddings of the object nodes are used to leverage their structural patterns in the knowledge source to compute similarity metrics for graph refinement and enrichment. We performed experimental and comparative analysis on the benchmark Visual Genome dataset, in which the proposed method achieved a higher recall rate ( $$R@K = 29.89, 35.4, 39.12$$ for $$K = 20, 50, 100$$ ) as compared to the existing state-of-the-art technique ( $$R@K = 25.8, 33.3, 37.8$$ for $$K = 20, 50, 100$$ ). The qualitative results of the proposed method in a downstream task of image generation showed that more realistic images are generated using the commonsense knowledge-based scene graphs. These results depict the effectiveness of commonsense knowledge infusion in improving the performance and expressiveness of scene graph generation for visual understanding and reasoning tasks.

Invariable of the agriculture type (precision, smart, or digital), the monitoring process of factors that increase the crop yield and growth is mostly non-ML, manually structured approaches with practical pain points. In this scenario, to reduce monitoring costs and maintenance efforts, there is a requirement for low-cost semi-autonomous distributed systems that can remotely collect plant data and perform standalone ML-based analytics without depending on cloud servers or the internet. In this work, we provide an embedded ML pipeline, which users can use/follow for end-to-end solution design and implementation for any of their use-cases. To demonstrate the pipeline, we use it to collect image data, train a CNN-based regression algorithm, perform hardware-specific tuning, generate optimized code, and deploy binaries on Sony Spresense setup. The initial testing shows that even the resource-constrained MCU-based Spresense, in real-time (992 ms), high performance (96.2% accuracy, 1.86 cm2RMSE), could analyze a plant in a semi-autonomous environment to predict the leaf area and plant growth.

Traditionally, original equipment manufacturers (OEMs) send device-specific over-the-air (OTA) packages to ensure the latest firmware, security patches, etc. With millions of IoT devices, even a tiny per-centage of OTA failures will result in tens of thousands of globally affected consumers. The state-of-the-art OTA methods are suited for high-end Android & embedded Linux devices and not for resource-constrained devices (e.g. MCUs, small CPUs) with only a few MB memory. Current OTA methods have been tested only on non-ML use-cases such as remote bugs patching or security flaws, etc. In this paper, we present OTA-TinyML approach that, via HTTPS, loads the C source file of ML models from a webserver into IoT boards. OTA-TinyML does not strain hardware (resource-friendly) as its implementation spans only a few lines of code. It is compatible with a range of ML models (e.g. text, speech, image domains) and MCUs (e.g. Cortex M series, STM32, Xtensa). OTA-TinyML is tested by performing remote fetching of 6 types of ML models, storing them on 4 types of memory units, then loading and executing on 7 popular Arduino IoT boards.

Smart Factories characterize as context-rich, fast-changing environments where heterogeneous hardware appliances are found beside of also heterogeneous software components deployed in (or directly interfacing with) IoT devices, as well as in on-premise mainframes, and on the Cloud. This inherent heterogeneity poses major challenges particularly when a high degree of resiliency is needed, and the ubiquitously deployed software components must be replaced or reconfigured at real-time to respond to the most diverse events, ranging from an out-of-range sensor detection, to a new order issued by a customer. In this work, a software framework is presented, which allows to deploy, (re)configure, run, and monitor the most diverse software across all the three layers of the Smart Factory (edge, fog, Cloud), from remote, via API calls, in a standardised uniform manner, relying on containerization technologies, and on a variety of software technologies, frameworks, and programming languages, including Node-RED, MQTT, Scala, Apache Spark, and Kafka. The most recent advances in the framework design, implementation, and demonstration, which led to the introduction of the so-called Crazy Nodes, are presented and motivated. A comprehensive proof-of-concept is given, where user interfaces and distributed systems are created from scratch via API calls to implement AI-based alerting systems, Big Data stream filtering and transformation, AI model training, storage, and usage for one-shot as well as stream predictions, and real-time Big Data visualization through line plots, histograms, and pie charts.

TinyML: Tiny in size, big in impact! In this paper, we present a Tensor Memory Mapping (TMM) method, which can accurately calculate the on-device execution memory consumed by a range of ML and TinyML models during execution on small central processing units (CPUs), microcontroller units (MCUs), and single board computers (SBCs).

Intelligent, smart, Cloud, reconfigurable manufac-turing, and remote monitoring, all intersect in modern industry and mark the path toward more efficient, effective, and sustain-able factories. Many obstacles are found along the path, including legacy machineries and technologies, security issues, and software that is often hard, slow, and expensive to adapt to face unforeseen challenges and needs in this fast-changing ecosystem. Light-weight, portable, loosely coupled, easily monitored, variegated software components, supporting Edge, Fog and Cloud computing, that can be (re)created, (re)configured and operated from remote through Web requests in a matter of milliseconds, and that rely on libraries of ready-to-use tasks also extendable from remote through sub-second Web requests, constitute a fertile technological ground on top of which fourth-generation industries can be built. In this demo it will be shown how starting from a completely virgin Docker Engine, it is possible to build, configure, destroy, rebuild, operate, exclusively from remote, exclusively via API calls, computation networks that are capable to (i) raise alerts based on configured thresholds or trained ML models, (ii) transform Big Data streams, (iii) produce and persist Big Datasets on the Cloud, (iv) train and persist ML models on the Cloud, (v) use trained models for one-shot or stream predictions, (vi) produce tabular visualizations, line plots, pie charts, histograms, at real-time, from Big Data streams. Also, it will be shown how easily such computation networks can be upgraded with new functionalities at real-time, from remote, via API calls.

Graphical User Interfaces (GUI) and real-time in-teractive Big Data charts play a key role in a wide variety of Big Data applications. The software libraries that are available at today are not suitable for displaying huge volumes of data in a single chart, because they require all the data to be collected at a single node. In this work, an innovative approach to the problem is presented, that consists in using a network of cyber-devices that is created and configured via API calls and that interfaces with a Scala Spark server application through a multiplicity of communication technologies, to produce and display a variety of time-space- infinite Big Data stream visualizations, including line plots, pie charts, histograms, that are updated at real-time as new data come, without ever collecting the data or the charts markup at a single node. The proposed approach characterizes for being (i) Web-based, (ii) API-based, (iii) Cloud-based, (iv) portable, (v) customizable/extendable, (vi) plug and play, and for relying on (i) Node-RED, (ii) MQTT, (iii) Scala, (iv) Akka HTTP, (v) Spark, (vi) Kafka, (vii) Docker. Remarkably, the same network used for Big Data visualization can be reconfigured in a matter of milliseconds and used for Big Data (streams) filtering, transformation, merge, analytics, and for training Machine Learning models, storing trained models on a Cloud storage, using stored models for one-shot or stream predictions, and much more. Although being at an advanced stage, we consider this research as a work in progress, since an extensive benchmarking and application to variegated real-world scenarios are still to be carried out.

The majority of IoT edge devices are embedded systems with a tiny microcontroller unit (MCU), which acts as its brain. When users want their edge devices to continuously improve for better edge-analytics results, there is a need to equip their devices with algorithms that can learn/train from the continuously evolving real-world data. Currently, such devices are not capable of executing any machine learning (ML)-based model training tasks due to their resource constraints such as: limited memory (SRAM, Flash, and EEPROM), low operations per second, its inability to perform parallel processing, etc. In this article, we provide ML-MCU, a framework with our novelOptimized-Stochastic Gradient Descent (Opt-SGD)andOptimized One-Versus-One (Opt-OVO)algorithms to enable both binary and multiclass ML classifier training directly on MCUs. Thus,ML-MCUenables billions of MCU-based IoT edge devices to self learn/train (offline) after their deployment, using live data from a wide range of IoT use cases. When evaluating our algorithms on multiple popular MCUs, using various data sets of different sizes and feature dimensions, one of the most exciting findings was, ourOpt-OVOalgorithm trained a multiclass classifier using a data set of class count 50, on a$\$ $3 resource-constrained MCU and also performed onboard unit inference for the same 50 class data in super real time (6.2 ms).

Electric vehicles (EVs) need to be recharged at intermediate locations, such as shopping malls, restaurants, and parking lots, to meet the daily commute requirements of their users. Currently, public electric vehicle supply equipment (EVSE) serve EVs by conventional methods, which can result in long waiting time for users. This issue reduces the travel efficiency of EVs and thus affects user comfort. Most previous research has studied energy consumption and charging cost optimization; however, comparatively less work has focused on waiting time optimization despite its great importance from the EV user’s perspective. In this paper, we formulate the waiting time optimization as a fuzzy integer linear programming problem and propose a novel heuristic fuzzy inference system-based algorithm (FISA) that resolves the objective function and minimizes the waiting time of EVs at public EVSE installations. We developed the underlying fuzzy inference system by defining the membership functions, expert rules, and formulation for obtaining the optimal solution. The novel FISA automates the correlations of the uncertain and independent input parameters into weighted control variables and resolves the objective function in each sampling period to optimize the waiting time for EVs with the most urgent service requirements. A java language-based simulator is developed for a parking lot to evaluate the effectiveness of the proposed FISA. The simulation results indicate higher efficiency of the proposed FISA compared with state-of-art scheduling algorithms.

Typically a Neural Networks (NN) is trained on data centers using historic datasets, then a C source file (model as a char array) of the trained model is generated and flashed on IoT devices. This standard process impedes the flexibility of billions of deployed ML-powered devices as they cannot learn unseen/fresh data patterns (static intelligence) and are impossible to adapt to dynamic scenarios. Currently, to address this issue, Online Machine Learning (OL) algorithms are deployed on IoT devices that provide devices the ability to locally re-train themselves -continuously updating the last few NN layers using unseen data patterns encountered after deployment.In OL, catastrophic forgetting is common when NNs are trained using non-stationary data distribution. The majority of recent work in the OL domain embraces the implicit assumption that the distribution of local training data is balanced. But the fact is, the sensor data streams in real-world IoT are severely imbalanced and temporally correlated. This paper introduces Imbal-OL, a resource-friendly technique that can be used as an OL plugin to balance the size of classes in a range of data streams. When Imbal-OL processed stream is used for OL, the models can adapt faster to changes in the stream while parallelly preventing catastrophic forgetting. Experimental evaluation of Imbal-OL using CIFAR datasets over ResNet-18 demonstrates its ability to deal with imperfect data streams, as it manages to produce high-quality models even under challenging learning settings.

The concept of ML model aggregation rather than data aggregation has gained much attention as it boosts pre- diction performance while maintaining stability and preserving privacy. In a non-ideal scenario, there are chances for a base model trained on a single device to make independent but complementary errors. To handle such cases, in this paper, we implement and release the code of 8 robust ML model combining methods that achieves reliable prediction results by combining numerous base models (trained on many devices) to form a central model that effectively limits errors, built-in randomness and uncertainties. We extensively test the model combining performance by performing 15 heterogeneous devices and 3 datasets based experiments that exemplifies how a complicated collective intelligence can be derived from numerous elementary intelligence learned by distributed, ubiquitous IoT devices.

Blockchain technology provides promising solutions to problems where trust between participants is a significant concern. Blockchain technology eliminates centralised trusted authorities and provides decentralised trust assessments and improved transparency and immutability for transactions. Blockchain platforms have tightly connected decentralised consensus mechanisms and distributed ledgers. However, within DApps(decentralised applications), blockchain platforms should easily communicate with other systems, including data storage systems and front-end applications. Blockchain is not for storing big data. The transactions cost is a primary concern when developing DApps for commercial use. This paper proposes a BaaS(Blockchain as a Service) architecture enriched with a loosely coupled service API(Application Programming Interface) connecting IPFS(Interplanetary File System) as the information storage and smart contracts to govern the business logic. We realised the proposed architecture for prescriptions management in a veterinary medicine case study in Ireland. The solution DApp continuously monitors the updates of prescriptions, medicine dispenses, and medicine administrations while restricting the overuse of prescription and the use of counterfeit drugs. The traceability improves the transparency of treatment activities and information retrieval for users. Moreover, the proposed architecture is easily adaptable and maintains low-cost transactions with the support of IPFS and smart contract events.

Transmitting updates of high-dimensional models between client IoT devices and the central aggregating server has always been a bottleneck in collaborative learning - especially in uncertain real-world IoT networks where congestion, latency, bandwidth issues are common. In this scenario, gradient quantization is an effective way to reduce bits count when transmitting each model update, but with a trade-off of having an elevated error floor due to higher variance of the stochastic gradients. In this paper, we propose ElastiCL, an elastic quantization strategy that achieves transmission efficiency plus a low error floor by dynamically altering the number of quantization levels during training on distributed IoT devices. Experiments on training ResNet-18, Vanilla CNN shows that ElastiCL can converge in much fewer transmitted bits than fixed quantization level, with little or no compromise on training and test accuracy.

Industry 4.0 is a recent trend of automation for manufacturing technologies and represents the fourth industrial revolution which transforms current industrial processes with the use of technologies such as automation, data analytics, cyber-physical systems, IoT, artificial intelligence, etc. The vision of Industry 4.0 is to build an end-to-end industrial transformation with the support of digitization. Data analytics plays a key role to get a better understanding of business processes and to design intelligent decision support systems. However, a key challenge faced by industry is to integrate multiple autonomous processes, machines and businesses to get an integrated view for data analytics activities. Another challenge is to develop methods and mechanisms for real-time data acquisition and analytics on-the-fly. In this paper, we propose a semantically interoperable framework for historical data analysis combined with real-time data acquisition, event detection, and real-time data analytics for very precise production forecasting within a manufacturing unit. Besides historical data analysis techniques, our middleware is capable of collecting data from diverse autonomous applications and operations in real time using various IoT devices, analyzing the collected data on the fly, and evaluating the impact of any detected unexpected events. Using semantic technologies we integrate multiple autonomous systems (e.g. production system, supply chain management and open data). The outcome of real-time data analytics is used in combination with machine learning models trained over historical data in order to precisely forecast production in a manufacturing unit in real time. We also present our key findings and challenges faced while deploying our solution in real industrial settings for a large manufacturing unit.

In the Infrastructure as a Service (IaaS) paradigm of cloud computing, computational resources are available for rent. Although it offers a cost efficient solution to virtual network requirements, low trust on the rented computational resources prevents users from using it. To reduce the cost, computational resources are shared, i.e., there exists multi-tenancy. As the communication channels and other computational resources are shared, it creates security and privacy issues. A user may not identify a trustworthy co-tenant as the users are anonymous. The user depends on the Cloud Provider (CP) to assign trustworthy co-tenants. But, it is in the CP's interest that it gets maximum utilization of its resources. Hence, it allows maximum co-tenancy irrespective of the behaviours of users. In this paper, we propose a robust reputation management mechanism that encourages the CPs in a federated cloud to differentiate between good and malicious users and assign resources in such a way that they do not share resources. We show the correctness and the efficiency of the proposed reputation management system using analytical and experimental analysis.

Distributed Social Networks (DSNs) are the solution to the privacy and security problems of online social networks. In DSN, a user controls their own data as it chooses personal storage for its social network data. In absence of a centralized entity with access to all social network data, information retrieval becomes difficult in DSNs. In this paper we propose to use crowd sourcing for information retrieval in a DSN. We analyze a popular information retrieval problem called expert search in a social network. In this paper, we present an algorithm for such a crowd sourcing based search process which includes solution for (a) the worker selection problem (b) the task selection problem and (c) the reward distribution problem. Using experimental evaluation, we show that, the search algorithms proposed in this paper can be as efficient as a greedy search algorithm with access to entire social network information.

User modeling of individual users on the Social Web platforms such as Twitter plays a significant role in providing personalized recommendations and filtering interesting information from social streams. Recently, researchers proposed the use of concepts (e.g., DBpedia entities) for representing user interests instead of word-based approaches, since Knowledge Bases such as DBpedia provide cross-domain background knowledge about concepts, and thus can be used for extending user interest profiles. Even so, not all concepts can be covered by a Knowledge Base, especially in the case of microblogging platforms such as Twitter where new concepts/topics emerge everyday. In this short paper, instead of using concepts alone, we propose using synsets from WordNet and concepts from DBpedia for representing user interests. We evaluate our proposed user modeling strategies by comparing them with other bag-of-concepts approaches. The results show that using synsets and concepts together for representing user interests improves the quality of user modeling significantly in the context of link recommendations on Twitter.

Opinion mining from customer reviews has become pervasive in recent years.Sentences in reviews, however, are usually classified independently, even though they form part of a review's argumentative structure.Intuitively, sentences in a review build and elaborate upon each other; knowledge of the review structure and sentential context should thus inform the classification of each sentence.We demonstrate this hypothesis for the task of aspect-based sentiment analysis by modeling the interdependencies of sentences in a review with a hierarchical bidirectional LSTM.We show that the hierarchical model outperforms two non-hierarchical baselines, obtains results competitive with the state-of-the-art, and outperforms the state-of-the-art on five multilingual, multi-domain datasets without any handengineered features or external resources.

In this paper, we study if reusing Google+ profiles can provide reliable recommendations on Twitter to resolve the cold start problem. Next, we investigate the impact of giving different weights for aggregating user profiles from two OSNs and present that giving a higher weight to the targeted OSN profile for aggregation allows the best performance in the context of a personalized link recommender system. Finally, we propose a user modeling strategy which combines entity-and category-based user profiles using with a discounting strategy. Results show that our proposed strategy improves the quality of user modeling significantly compared to the baseline method.

The main contribution of this work is the comparison of three user modeling strategies based on job titles, educational fields and skills in LinkedIn profiles, for personalized MOOC recommendations in a cold start situation. Results show that the skill-based user modeling strategy performs best, followed by the job- and edu-based strategies.

Citizen Actuation is a new concept that aims to retain humans in the loop throughout a system's lifecycle. In system design, humans are (generally) just users of a system but both Citizen Sensing and Citizen Actuation rely on users being included in a Cyber Physical Social System. In this paper, we investigate employing profile features from social networks as a method for user selection. These users will then be sent small tasks to complete that might normally be undertaken by actuators. To achieve this, we conducted a survey where users evaluated profiles on a limited number of features and posts. Separately, we collected profile data from the same set of profiles and computed calculated values such as Reply Ratio to compare them with the survey findings. This study has revealed interesting insights in to what the survey participants find important in relation to social media profiles and completing tasks. These include insights such as how they view the number of tweets, the profile description text, and how a user interacts with other users as being important when forming an opinion on a profile.

The logging of Activities of Daily Living (ADLs) is becoming increasingly popular mainly thanks to wearable devices. Currently, most sensors used for ADLs logging are queried and filtered mainly by location and time. However, in an Internet of Things future, a query will return a large amount of sensor data. Therefore, existing approaches will not be feasible because of resource constraints and performance issues. Hence more fine-grained queries will be necessary. We propose to filter on the likelihood that a sensor is relevant for the currently sensed activity. Our aim is to improve system efficiency by reducing the amount of data to query, store and process by identifying which sensors are relevant for different activities during the ADLs logging by relying on Distributional Semantics over public text corpora and unsupervised hierarchical clustering. We have evaluated our system over a public dataset for activity recognition and compared our clusters of sensors with the sensors involved in the logging of manually-annotated activities. Our results show an average precision of 89% and an overall accuracy of 69%, thus outperforming the state of the art by 5% and 32% respectively. To support the uptake of our approach and to allow replication of our experiments, a Web service has been developed and open sourced.

Websites that provide content creation and sharing features have become quite popular recently. These sites allow users to categorize and browse content using "tags” or free-text keyword topics. Since users contribute and tag social media content across a variety of social web platforms, creating new knowledge from distributed tag data has become a matter of performing various tasks, including publishing, aggregating, integrating, and republishing tag data. In this paper, we introduce an object-centered social network based on tagging practices across different sources, and then we show how this network can be built and emerged over time.

Social tagging in online communities has become an important method for reflecting classified thoughts of individual users. A number of social Web sites provide tagging functionalities and also offer folksonomies within or across the sites. However, it is practically not easy to find users' interests based on such folksonomies. In this paper, we provide a novel approach for clustering user-centric interests by analyzing tagging practices of individual users. To do this, we collect Really Simple Syndication data from blogosphere, find conceptual clusters using formal concept analysis, and then evaluate the significance of these clusters. The results of the empirical evaluation show that we can effectively recommend different collections of tags to an individual or a set of users.

Argumentative discussions are common in Web 2.0 applications, but the social Web still offers limited or no explicit support for argumentation. As Web 2.0 applications become more popular, modeling argumentation happening in these systems becomes important, to enable reuse and further understanding of online discussions. After reviewing four genres of online conversations – Web bulletin boards, Wiki talk pages, blog comments, and microblogs – and four current Web 2.0 argumentation systems, the paper suggests how Semantic Web technologies can be used to provide an interoperability layer for argumentation modeling across applications.

While the use of Web 2.0 tools and principles in organizations - a practice commonly known as Enterprise 2.0 - helps knowledge workers to collaboratively build and exchange information more easily, it introduces several issues in terms of efficiently integrating and retrieving this information. In t

Although tagging is a widely accepted practice on the Social Web, it raises various issues like tags ambiguity and heterogeneity, as well as the lack of organization between tags. We believe that Semantic Web technologies can help solve many of these issues, especially considering the use of formal resources from the Web of Data in support of existing tagging systems and practices. In this article, we present the MOAT—Meaning Of A Tag—ontology and framework, which aims to achieve this goal. We will detail some motivations and benefits of the approach, both in an Enterprise 2.0 ecosystem and on the Web. As we will detail, our proposal is twofold: It helps solve the problems mentioned previously, and weaves user-generated content into the Web of Data, making it more efficiently interoperable and retrievable.

Publishing information on the Semantic Web using common formats enables data to be linked together, integrated and reused. In order to fully leverage the potential for interlinking data by reusing existing schemas, an intuitive way of viewing current usage of RDF vocabularies is required. We present a system which allows a user to view the most frequently occurring namespaces and classes in a large Semantic Web dataset, and the main linkage patterns that exist between them. Users can select a namespace of interest in order to examine usage of a particular ontology, and see how it is being combined with other vocabularies.

Classic argumentative discussions can be found in a variety of domains from traditional scientific publishing to today's modern social software. An interactive argumentative discussion usually consists of an initial proposition stated by a single creator, followed by supporting propositions or counter-propositions from other contributors. Thus, the actual argumentation semantics is hidden in the content created by the contributors. Although there are approaches that try to deal with this challenge, most of them focus on a particular domain, limiting the scope of the argumentation to that domain only. In this paper, we describe an abstract model for argumentation which captures the semantics independently of the domain. Following a modularized approach, we also take into account additional important aspects of the argumentation, like the provenance information or its evolution (the temporal side).

Wikis are popular collaborative hypertext authoring environments, but they neither support structured access nor information reuse. Adding semantic annotations helps to address these limitations. We present an architecture for Semantic Wikis and discuss design decisions including structured access, views, and annotation language. We present our prototype SemperWiki that implements this architecture.