Accurate weather prediction remains a persistent challenge due to the multi-component nature of atmospheric systems and the limitations of traditional forecasting models. This study presents an Artificial Intelligence (AI)–driven weather forecasting framework that integrates Machine Learning (ML), Deep Learning (DL), self-supervised learning, ensemble learning, and explainable AI techniques. Particularly, we propose a Self-Supervised Multi-Fidelity Ensemble Learning (SSMFEL) model, which integrates low-, mid-, and high-fidelity long short-term memory architectures with regularization, dropout, and a decoder to extract meaningful patterns from unlabeled data. The proposed model achieved over 95% R-squared ( R 2 ) metric, coefficient of determination, for humidity and temperature prediction and more than 52% R 2 for wind speed and precipitation, significantly outperforming existing ML, DL, and transformer-based models. SSMFEL also demonstrated the 0.00018 Mean Squared Error (MSE) and 0.03489 root MSE across all components. Results from test statistic tests , sensitivity analysis, and fairness evaluation further confirmed the statistical superiority of the proposed approach. Model explainability analysis revealed that specific humidity is the most influential feature, while temperature and relative humidity contribute contextually. Theoretical analysis based on bias–variance reduction and generalization theory further supports the improved forecasting performance of the proposed method.

Machine learning (ML) has become integral to cybersecurity applications, e.g., phishing detection, intrusion detection systems, malware analysis, and botnet identification. However, the integration of ML also exposes novel attack surfaces that can be exploited through adversarial machine learning (AML). While prior surveys have examined individual threats or defenses, they often focus narrowly on specific stages, e.g., training or testing. In contrast, in this paper, we provide the first comprehensive survey of adversarial attacks and defenses across the entire ML development life cycle within the cybersecurity domain. Using a structured methodology, we categorize vulnerabilities and countermeasures at each stage, data gathering, model training, testing, deployment, and maintenance, highlighting cross-stage interactions and emerging distributed threat models. Our study addresses key gaps in current defenses, including their limited generalizability and lack of standardized evaluation practices, and identifies promising directions, e.g., lifecycle-aware robustness, distributed resilience, and the integration of statistical with generative methods. Consolidating fragmented research into an end-to-end perspective, this study advances the understanding of AML in cybersecurity and outlines a roadmap for building more trustworthy, and resilient ML-driven security systems.

Edge computing enables real-time Internet of Things data processing by bringing computation closer to data sources, but its distributed architecture creates cybersecurity vulnerabilities requiring privacy-preserving attack detection mechanisms capable of handling heterogeneous data distributions. This article proposes federated generative adversarial divergence (FedGAD), a plug-and-play modular framework that enhances existing federated learning methods through Jacobian-based regularization and dynamic complexity-aware weighting to address cyber-attack detection in non-independent and identically distributed (IID) and unlabeled edge data environments. Unlike existing approaches suffering from mode collapse and training instability, FedGAD maintains statistical consistency across distributed nodes through gradient-based stability mechanisms, supported by rigorous theoretical analysis establishing convergence guarantees and mode coverage properties. We conduct comprehensive experiments comparing FedGAD against four federated generative learning baselines federated trustworthy (FedTrust), anomaly detection generative adversarial network (ADGAN), federated generative adversarial network for intrusion detection system (FedGAN-IDS), and federated temporal sequential recurrent generative network (FedTSRGNet) and four regularization-based methods federated averaging (FedAvg), federated proximal (FedProx), learning with collaborative aggregation method (LeCam), and Jensen Shannon (JS) Divergence on telemetry data of networks - internet of things (ToN_IoT) and Communications Security Establishment in Canadian Institute for Cybersecurity - Intrusion Detection System (CSE_CIC_IDS) datasets, demonstrating FedGAD's superiority with accuracy improvements up to 3.5%, achieving 100% mode coverage compared to 25% for baseline methods while maintaining computational efficiency for resource-constrained edge deployments.

Insider threats pose significant security risks in distributed networks, because employees within the organisation may misuse their access to compromise systems. Centralised Machine Learning (ML) techniques are inappropriate in these situations due to privacy and data heterogeneity concerns. To address class imbalance and non-IID data, this study introduces FedAT, a Federated Adversarial Training that integrates federated learning (FL) with generative models to deliver privacy-preserving, multiclass Insider Threat Detection (ITD). FedAT outperforms centralized and conventional FL techniques in terms of scalability, privacy preservation, and detection accuracy, according to evaluations conducted on public CERT datasets.

Industrial Cyber-Physical Systems (ICPSs) have been motivating various Industry 4.0 endeavours, particularly with the integration of fog computing. Cloud-fog data caching paradigms, as supportive elements of ICPSs, have been adopted to cache user data, catering to diverse ICPS requirements such as data sensitivity and reduced access latency. In this hierarchical caching context, ensuring Cloud-Fog Data Integrity (CFDI) is crucial for maintaining the consistent functionality of ICPSs. Existing solutions primarily focus on examining the integrity of data cached solely on either cloud or fog nodes. However, cloud-cached data and fog-cached data are tightly coupled and should be considered simultaneously when checking data integrity. In this work, we introduce an over-the-air CFDI verification scheme, namely AirDIV, with a high accuracy and security guarantee. Instead of aggregating integrity proofs after proof transmission, AirDIV completes proof aggregation and transmission over the air for efficiency improvement. To enhance practicability, we derive adjustable parameters and formulate an optimization problem to minimize over-the-air aggregation errors. Furthermore, with an effective proof generation method, AirDIV can defend against two common attacks, i.e., replay and forge attacks. We provide a theoretical analysis of AirDIV’s correctness, accuracy and security, while conducting extensive experiments on both simulated and real platforms to validate its efficiency.

Knowledge Graph Embedding (KGE), which automatically capture structural information from Knowledge Graphs (KGs), are essential for enhancing various downstream tasks, such as recommender systems. To further improve the effectiveness of KGE models, Federated Knowledge Graph Embedding (FKGE) has been introduced, enabling the privacy-preserving integration of KGs across multiple organizations. However, existing FKGE frameworks require aggregation of a large global KGE model (embeddings). resulting in significant communication overhead, thereby reducing the efficiency and utility of FKGE in practical scenarios. To address this challenge, we propose Federated Compositional Knowledge Graph Embedding (FedComp), which enhances communication efficiency by leveraging the compositional characteristics of KG entities. In FedComp, we design a lightweight global model that represents shareable latent features of entities. These global latent features are composed into personalized KGE models with local embedding generators on the clients, improving both local adaptability and performance. By this, FedComp can significantly reduce the number of parameters that need to be transmitted Experimental results show that FedComp outperforms state-of-the-art FKGE frameworks on link prediction accuracy, with only around 1.0% communication overhead compared to counterpart frameworks.

Edge intelligence is an emerging distributed computing paradigm that has been driven by the rapid proliferation of Internet of Things (IoT) devices, along with the advancements in edge computing and artificial intelligence. With latency-sensitive data commonly cached across multiple Edge Servers (ESs), efficient Edge Data Integrity Verification (EDIV) has become increasingly critical. Traditional ‘challenge-response’ EDIV methods incur substantial computation and communication costs by indiscriminately verifying all ESs, even though not all ESs may be simultaneously corrupted. A recent Federated Learning (FL)-based framework partially addressed this inefficiency by identifying potentially corrupted ESs early, considering only homogeneous ES activity data. However, due to heterogeneous activity data across diverse ESs, this approach suffers from reduced detection accuracy of potentially corrupted ESs, slower FL convergence, and unclear guidance for subsequent verification rounds, thus limiting the overall reduction in EDIV computation and communication costs. To that end, we proposeFederated learning withAdaptiveRegularizer-basedEdgeDataIntegrityVerification (FedAR-EDIV), which is an effective FL-based framework integrating an adaptive objective regularization strategy specifically designed to handle heterogeneous data distributions. FedAR-EDIV efficiently identifies potentially corrupted ESs during the FL process, achieves faster convergence, and significantly reduces computation and communication costs in the final EDIV procedure. It achieves up to 16× communication speedup and 9.1× computation cost reduction compared to baseline EDIV methods, and reaches FL-based detection accuracy exceeding 99.78% under heterogeneous conditions using KDD99 activity data. Additionally, FedAR-EDIV incorporates a dynamic reputation mechanism after each EDIV round to strategically guide subsequent verification rounds, ensuring fewer checks for trustworthy ESs and greater scrutiny for suspicious ones, thus further minimizing EDIV-related costs. We provide a theoretical analysis that demonstrates the convergence of FedAR-EDIV during FL training, as well as correctness, efficiency, and security during the EDIV process. Extensive experiments conducted on two different heterogeneous activity datasets validated that FedAR-EDIV substantially outperforms baseline methods in terms of corrupted ES detection accuracy, FL convergence speed, and overall EDIV computation and communication costs.

Federated Learning (FL) allows collaborative training of a Machine Learning (ML) model while preserving data privacy across participating clients. Most existing studies consider FL clients to be proactive and completely honest in their participation. However, in reality, clients might lack the motivation to participate, and malicious behavior among some clients could negatively impact the interests of others. For these reasons, ensuring trust and fairness among FL clients is paramount but remains challenging due to limitations in FL consensus mechanisms and incentive strategies. To address these challenges, we introduce a Trustworthy and Fair FL (TFFL) framework that develops a reputation-based consensus mechanism called Dynamic Reputation Consensus (DRC), where clients’ reputations are dynamically assessed based on subjective opinions by evaluating real-time client behavior. We also incorporate time decay and temporal discounting of TFFL interactions along with the weighted measures of clients’ data quality, performance, and reliability to accurately reflect the evolving nature of client behavior over time. By adaptively adjusting clients’ incentives based on reputations and a cooperative game theory, DRC incentivizes honest participation and discourages malicious intent. In addition, we utilize blockchain and smart contracts to provide decentralized, regularized, and secure reputation management that is resistant to tampering and non-repudiation. Theoretical analysis and empirical results on widely used datasets (MNIST, CIFAR-10, and CIFAR-100) demonstrate the effectiveness of DRC in enhancing trust and fairness, improving performance, and providing robust security in FL settings. Results further exhibit that DRC offers superior performance in local model validation, consensus decision, and convergence time compared to related research approaches across various experimental settings.

Ensuring Edge Data Integrity (EDI) is imperative in providing reliable and low-latency services in mobile edge computing. Existing EDI schemes typically address single-vendor (App Vendor, AV) single-server (Edge Server, ES), single-vendor multi-server, and multi-vendor multi-server scenarios, which consider a single data replica cached by an ES from the AVs. However, the most practical scenario of Multi-Vendors and Multi-Servers with Multiple Data (MVMS-MD) cached by an ES from different AVs remains unexplored. Current solutions struggle when applied to this scenario due to increased computation and communication costs in the verification process across all ESs using the classicalchallenge-response per-data multi-roundstrategy. To tackle this issue, we propose a Multiple EDI-Verification (MEDI-V) approach in this paper. In particular, our MEDI-V utilizes an adaptive Merkle Hash Tree (ad-MHT) to efficiently generate a tree of multiple data replicas within each AV. Next, the dynamic mechanism computes minimal verification information using ad-MHT to create achallengefor individual ESs to produce EDI proofs. The ES then leverages its ad-MHT and the ES's proof to send the reconstructed ad-MHT root to the AV for verification. Theoretical insights into MEDI-V's correctness, efficiency, security, and comprehensive evaluations demonstrate its superiority in addressing MEDI issues in the MVMS-MD scenario.

Federated Learning (FL) is a machine learning technique that enables multiple local clients holding individual datasets to collaboratively train a model, without exchanging the clients' datasets. Conventional FL approaches often assign a fixed workload (local epoch) and step size (learning rate) to the clients during the client-side local model training and utilize all collaborating trained models' parameters evenly during the server-side global model aggregation. Consequently, they frequently experience problems with data heterogeneity and high communication costs. In this paper, we propose a novel FL approach to mitigate the above problems. On the client side, we propose an adaptive model update approach that optimally allocates a needful number of local epochs and dynamically adjusts the learning rate to train the local model and regularizes the conventional objective function by adding a proximal term to it. On the server side, we propose a robust model aggregation strategy that potentially supplants the local outlier updates (models' weights) prior to the aggregation. We provide the theoretical convergence results and perform extensive experiments on different data setups over the MNIST, CIFAR-10, and Shakespeare datasets, which manifest that our FL scheme surpasses the baselines in terms of communication speedup, test-set performance, and global convergence.

Federated Learning (FL) is an emerging research area that produces a globally trained model using numerous local users' data and maintains their privacy. Heterogeneous or non-Independent and Identically Distributed ( non-IID) data affect the global model's convergence and, therefore, cause high communication costs. These are because traditional FL approaches often disregard an adaptive regularized objective for the user-side training and utilize conventional arithmetic mean on the locally trained models for the server-side aggregation. To alleviate these issues, we propose a novel FL scheme in this paper. In particular, we propose an adaptive regularization approach to add to the classical objective function of the users' local models during training and a resilient estimation approach to the locally trained models during aggregation. The adaptive regularization approach is derived using the users' local and global performance diversification while the resilient estimation scheme uses a modified geometric mean aggregation over the local models' parameters. We provide consolidated theoretical results and perform extensive experiments on the IID and non-IID settings of MNIST, CIFAR-10, and Shakespeare datasets with various deep networks. The results manifest that our FL scheme outperforms the state-of-the-art approaches in terms of communication speedup, test-set performance, training convergence stability, and resiliency against attacks.

MobileEdgeComputing (MEC) is an emerging technology, where App vendors are allowed to cache multiple data replicas on geographically distributed edge servers to serve adjacent mobile subscribers. However, this benefit introduces an extra workload for edge servers and App vendors, as they must audit the integrity of multiple data replicas periodically considering various threats caused by distributed and dynamic MEC environments. The large-scale growth of data replicas certainly is a challenge to design more efficientEdgeDataIntegrity (EDI) verification approaches. Existing solutions are mostly limited to improving efficiency by optimizing proof generation and verification methods, while the improvement is still far from satisfactory due to adopting indiscriminate inspection philosophy (checking all data replicas without discrimination). In this paper, we make the first attempt to abstract a pre-processing phase and correspondingly study theUnreliable dataReplicaSelection (URS) problem. It can be seamlessly integrated into existing EDI solutions by solving the URS problem at the start of each verification round. Such pre-selection can significantly enhance overall EDI verification efficiency by incorporating the cache serviceQualityofService (QoS) and verification success rate, especially in scenarios with a large number of data replicas. Specifically, we first formalize the URS problem as a constrained optimization problem, and further prove its$\mathcal {NP}$-hardness. To address the problem efficiently, we transform it into an easy-to-handle form and develop aPriority-based approach named URS-P. Both theoretical analysis and experimental evaluation validate the effectiveness and efficiency of our proposed solution.

Abstract Educational Data Mining (EDM) has become a promising research field for improving the quality of students and the education system. Although EDM dates back to several years, there is still lack of works for measuring and enhancing the computer programming skills of tertiary students. As such, we, in this paper, propose an EDM system for evaluating and improving tertiary students’ programming skills. The proposed EDM system comprises two key modules for (i) classification process and (ii) learning process,. The classification module predicts the current status of a student and the learning process module helps generate respective suggestions and feedback to enhance the student’s quality. In particular, for the classification module, we prepare a real dataset related to this task and evaluate the dataset to investigate six key Machine Learning (ML) algorithms, Support Vector Machine (SVM), decision tree, artificial neural network, Random Forest (RF), k-nearest neighbor and naive Bayes classifier, using accuracy-related performance measure metrics and goodness of the fit. The experimental results manifest that RF and SVM can predict the students more accurately than the other models. In addition, critical factors analysis is accomplished to identify the critical features toward achieving high classification accuracy. At last, we design an improvement mechanism in the learning process module that helps the students enhance their programming skills.

Existing Federated Learning (FL) algorithms generally suffer from high communication costs and data heterogeneity due to the use of conventional loss function for local model update and the equal consideration of each local model for global model aggregation. In this paper, we propose a novel FL approach to address the above issues. For local model update, we propose a disentangled Information Bottleneck (IB) principle-based loss function. For global model aggregation, we suggest a model selection strategy based on Mutual Information (MI). Particularly, we design a Lagrangian-based loss function using the IB principle and “disentanglement” for maximizing MI between the ground truth and model prediction and minimizing MI between the intermediate representations. We calculate MI ratio between the ground truth and model prediction, and between the original input and ground truth to select the effective models for aggregation. We analyze the theoretical optimal cost of the loss function and manifest optimal convergence rate, and quantify the outlier robustness of the aggregation scheme. Experiments demonstrate the superiority of the proposed FL approach, in terms of testing performance and communication speedup (i.e., 3.00-14.88 times for IID MNIST, 2.5-50.75 times for non-IID MNIST, 1.87-18.40 times for IID CIFAR-10, and 1.24-2.10 times for non-IID MIMIC-III).

Federated Averaging (FedAvg) is the baseline Federated Learning (FL) algorithm that applies the stochastic gradient descent for local model training and the arithmetic averaging of the local models’ parameters for global model aggregation. Succeeding FL works commonly utilize the arithmetic averaging scheme of FedAvg for the aggregation. However, such arithmetic averaging is prone to the outlier model-updates, especially when the clients’ data are non-Independent and Identically Distributed (non-IID). As such, the classical aggregation approach suffers from the dominance of the outlier updates and, consequently, causes high communication costs towards producing a decent global model. In this letter, we propose a robust aggregation strategy to alleviate the above issues. In particular, we propose first pruning the node-wise outlier updates (weights) from the local trained models and then performing the aggregation on the selected effective weights-set at each node. We provide the theoretical result of our method and conduct extensive experiments on the MNIST, CIFAR-10, and Shakespeare datasets with IID and non-IID settings, which demonstrate that our aggregation approach outperforms the state-of-the-art methods in terms of communication speedup, test-set performance and training convergence.

Federated Learning (FL) is an emerging research field that yields a global trained model from different local clients without violating data privacy. Existing FL techniques often ignore the effective distinction between local models and the aggregated global model when doing the client-side weight update, as well as the distinction of local models for the server-side aggregation. In this article, we propose a novel FL approach with resorting to mutual information (MI). Specifically, in client-side, the weight update is reformulated through minimizing the MI between local and aggregated models and employing Negative Correlation Learning (NCL) strategy. In server-side, we select top effective models for aggregation based on the MI between an individual local model and its previous aggregated model. We also theoretically prove the convergence of our algorithm. Experiments conducted on MNIST, CIFAR-10, ImageNet, and the clinical MIMIC-III datasets manifest that our method outperforms the state-of-the-art techniques in terms of both communication and testing performance.