Search Results (364)

AI technologies are becoming operational components of urban infrastructure systems, including transport networks, structural health monitoring platforms, water utilities, energy systems, and public facilities. These systems support prediction, diagnosis, control, maintenance planning, and asset-management decisions across long service periods. However, much of the existing literature still evaluates infrastructure AI at the model-design or deployment-performance stage, with limited attention to post-deployment validity, operational degradation, update control, and end-of-life management. This survey examines AI applications in urban infrastructure from a lifecycle-management perspective, covering deployment, runtime monitoring, maintenance and adaptation, governance, and retirement. The review applies a PRISMA-guided search and screening protocol to classify retained studies by lifecycle phase, infrastructure domain, deployment evidence, monitoring strategy, adaptation mechanism, governance control, and benchmark support. The cross-domain analysis indicates that deployment-stage accuracy alone is not sufficient for long-term reliability assessment, because sensor wear, environmental variation, asset aging, data drift, maintenance intervention, topology change, and operating-regime shifts can alter model behavior after deployment. The findings further show that current research provides limited support for linking model outputs to maintenance actions, validating model updates under operational constraints, documenting governance evidence, estimating lifecycle cost, defining retirement criteria, and building shared lifecycle benchmarks. The survey concludes that urban infrastructure AI should be managed as a long-term socio-technical asset, with continuous validation, model-health monitoring, controlled adaptation, audit-ready governance, and retirement planning integrated into infrastructure operations. Full article

Existing platform-monitoring methods for sport communities commonly rely on isolated descriptive text-mining outputs or general sentiment scores; they rarely integrate interpretable ESG issue coding with class-sensitive risk detection and provide limited support for auditable, privacy-conscious analysis of safety, ethics, and institutional trust. These limitations motivate a multi-stage framework that converts heterogeneous platform discourse into complementary structural and evaluative signals. Conceptually, digital trust is treated as the focal governance outcome; ethics and safety are substantive domains of concern; ESG provides the bounded classification and response ontology; and early warning denotes a prototype, human-reviewed weak-signal triage concept rather than incident prediction or a deployed security-monitoring system. Using 377,700 cleaned Korean-language comments on judo and Brazilian Jiu-Jitsu (BJJ) collected from Naver News and YouTube between 2010 and 2025, the framework combines n-gram analysis, LDA topic modeling, CONCOR network analysis, bounded ESG discourse classification, and three-class sentiment prediction. The individual analytical algorithms are established; the methodological contribution lies in their governance-oriented orchestration through a bounded ESG/non-ESG coding gate, a study-specific index layer, and a human-reviewed pathway from aggregate discourse signals to proportionate review. The analytical workflow identifies issue salience, relational topic structures, ESG dimensions, sentiment risk, legitimacy balance, and platform-specific risk concentration while excluding personally identifiable information. Empirically, social and governance concerns dominate the corpus, and governance-related negative sentiment consistently exceeds social-risk sentiment, highlighting rule transparency, coach ethics, misinformation, platform reputation, and institutional response as central trust-risk domains. Cell-weighted sensitivity checks preserved the governance-over-social and YouTube-over-Naver risk ordering, although the pooled salience estimate remained sensitive to the rapid expansion of BJJ discourse on YouTube. The fine-tuned KLUE-BERT model achieved a Macro-F1 of 0.838 and a negative-class F1 of 0.862, outperforming the strongest baseline, Text-CNN (Macro-F1 = 0.791), by 0.047 absolute Macro-F1 points (approximately 6.0% relative improvement). These findings support the feasibility of a batch-oriented, human-reviewed prototype for prioritizing aggregate discourse patterns. They do not establish the effectiveness of a real-time security-monitoring, incident-detection, or operational early-warning system. Full article

Temporal super-resolution (SR) aims to reconstruct a high-resolution signal from a low-resolution observation. When hardware limits force low sampling rates, this problem becomes critical for non-stationary signals with abrupt transients and rapid spectral changes. This manuscript reports a deterministic case study using pretrained 1D convolutional models and deterministic evaluation on paired real, synthetic, and mixed EEG-like signals. A compact encoder–linear upsampler–refinement architecture is evaluated at 5× upsampling under four training regimes: synthetic-only, real-only, tuned-real, and mixed. Performance is assessed with Mean Squared Error (MSE), Mean Absolute Error (MAE), Normalized Mean Absolute Error (NMAE), Log Spectral Distance (LSD), and spectral correlation (SCORR). Across 12 model–dataset combinations, mixed-domain training yields the most robust cross-domain behavior, outperforming single-domain checkpoints on real and mixed evaluation subsets. These findings support the practical value of training corpus composition for temporal SR under distribution shift. A focused morphological event analysis further shows that reconstruction error concentrates at abrupt amplitude and frequency boundaries, confirming that these transient regions are the dominant local challenge. An exploratory hybrid wavelet–superlet pilot is also reported; it achieves competitive pointwise error on selected domains but exhibits a substantial spectral-fidelity gap, indicating that frequency-aware inputs alone do not guarantee spectral reconstruction without auxiliary spectral losses. Full article

This paper gives a compact, constructive method for splitting a broad class of Eisenstein–Jacobi (EJ) interconnection network graphs into three edge-disjoint Hamiltonian cycles. EJ networks are hexagonal quotient-lattice graphs with three natural direction classes; a six-regular non-axis EJ network can contain at most three such cycles, so this decomposition is optimal. The method starts from the natural direction factors, applies elementary unit-parallelogram switches to the first two, and proves that the remaining edges form the third Hamiltonian cycle. Within the canonical ordered local-switch model, $d=1$ needs no switches, $d=2$ has optimal total cost four, and $d=3$ and $d\ge 4$ attain the component-counting lower bound in each modified factor. For $d\ge 4$, a two-valued alternating lift cancels the reduced-ratio dependence in the fine diagonal coordinate. A fine-incidence rank certificate then proves complement connectivity by showing that all diagonal arcs and released connectors form one cycle of length $4d-6$. The compact certificate contains $O\left(d\right)$ seed records and expands to the full $O\left(N\right)$ edge lists only when required. Deterministic symbolic, full-quotient, and dictionary-free incidence audits are supplied as reproducibility files. Full article

Service-Based Architectures (SBAs) in 5G core and cloud-native deployments require differentiated treatment for service classes with heterogeneous latency, reliability, and throughput expectations. Although HTTP/3 over QUIC is an important evolution of the HTTP ecosystem, HTTP/2 remains operationally relevant in SBA environments where TCP/TLS-based infrastructures and 3GPP service-based interfaces continue to rely on HTTP/2 communication. This paper therefore focuses on HTTP/2 priority signaling and the problem of translating application-level Service Level Agreement (SLA) policies into protocol-level priority metadata. To address this problem, the paper presents an SLA-aware priority management system built around the RFC 9218 extensible prioritization scheme, specifically its urgency and incremental parameters. The system integrates three coordinated subsystems: a rule-based Priority Classification Engine (PCE), a feedback-driven Dynamic Priority Mapping Algorithm (DPMA), and a runtime priority-update manager that applies bounded priority adjustments under changing network and load conditions. The revised evaluation reports a 7200-observation baseline campaign covering four operating modes, ten service classes, nine network profiles, and twenty repetitions per service–profile–mode combination, together with a 14,880-observation scalability and overhead campaign across increasing concurrent-stream levels. Compared with the unmanaged HTTP/2 baseline, DPMA reduced mean latency by 24.8%, P95 latency by 35.1%, P99 latency by 38.0%, and SLA violations by 19.9 percentage points. Compared with the legacy RFC 7540 baseline, DPMA reduced mean latency by 39.0%, P95 latency by 49.3%, P99 latency by 49.9%, and SLA violations by 21.1 percentage points. Compared with the static RFC 9218 baseline, DPMA reduced mean latency by 38.7%, P95 latency by 48.1%, P99 latency by 50.6%, and SLA violations by 21.4 percentage points. The scalability analysis shows that DPMA maintained P95 latency between 126.8 ms and 128.2 ms over the tested 1–100 concurrent-stream range, with priority-update decision overhead below 0.004 ms per request. The results indicate that SLA-aware use of RFC 9218 priority metadata can improve latency and SLA-compliance behavior in controlled SBA-like HTTP/2 environments while preserving a transparent and auditable prioritization policy. Full article

With the rapid development of Vehicular Ad Hoc Networks (VANETs), Vehicle-to-Infrastructure (V2I) communication plays a critical role in Intelligent Transportation Systems (ITS). However, existing authentication and key exchange protocols face challenges such as high computational cost, large communication overhead, and security and privacy risks in high-speed mobile environments. To address these problems, this paper proposes a lightweight V2I authentication key exchange and ticket-based fast handover authentication protocol based on consortium blockchain and a Physical Unclonable Function (PUF). The proposed framework integrates PUF-based device binding, biometric-assisted user binding, PRF-based dynamic pseudonym update, target-RSU-bound handover tickets, and consortium blockchain-assisted auditability. To avoid privacy leakage on immutable ledgers, the blockchain stores only keyed pseudonym indexes, cryptographic commitments, timestamps, revocation states, and audit records, whereas biometric helper information, PUF-derived values, long-term secrets, handover keys, and session keys are protected in TPM/HSM or encrypted off-chain storage. Formal verification using ProVerif indicates that the revised protocol satisfies the modeled secrecy properties, injective mutual authentication for initial authentication and handover, and non-injective ticket origin authenticity for accepted handover tickets. In addition, the Real-or-Random (RoR) model is used to prove fresh session key indistinguishability under explicit pre- and post-Test freshness, PUF unpredictability, fuzzy extractor security, and hardware-protected secret assumptions. Analytical performance evaluation further shows the core cryptographic cost of the proposed scheme while explicitly separating and parameterizing deployment-dependent TPM/HSM, AEAD, blockchain lookup, and PBFT confirmation costs. Full article

Stadium stands combine high occupant density, steep seating geometry, and constrained egress routes, yet design-stage egress calculations usually represent spectators through aggregate flow rates and do not explicitly account for health-related stair-use tolerance. This study proposes a reproducible queue-based workflow for estimating health-risk-induced egress delay in independent stadium-stand units and for supporting early-stage design screening. A deterministic capacity–flow baseline is first computed for each stand and translated into an auditable GraphML queue network. A Nursing-Based Health Score (NBHS), derived from riser height, tread depth, and stair pitch, is used as a geometry-based tolerance proxy rather than a clinical diagnostic score. NBHS risk is then propagated through vulnerable-profile headway multipliers at constrained gangway-to-vomitory bottleneck edges, and stand-level delay is estimated using grouped Monte Carlo queue processing. The queue baseline remained close to the deterministic reference (mean $T_{queue0}$/$T_{det}$ = 0.980), supporting interpretation of subsequent delays as controlled perturbations. Delay increased monotonically with headway sensitivity, and stand-level variation was explained by GraphML-derived bottleneck service-rate properties. These findings provide a practical design-stage reference for identifying stand configurations where stair geometry and egress bottleneck capacity jointly amplify delay for vulnerable spectators. Full article

This paper proposes a federated learning framework for privacy-preserving patient monitoring with lightweight blockchain anchoring. The framework keeps synthetic patient monitoring records local at each client and uses federated model aggregation to support collaborative learning without centralizing raw records. To improve traceability, the blockchain layer is specified as an anchoring mechanism that records compact evidence, including model hashes and participation metadata, rather than raw data or full model parameters. Experiments were conducted on synthetic patient monitoring data to evaluate framework behavior under non-IID client distributions, label noise, different client counts, partial client participation, and aggregation strategies. The centralized MLP baseline achieved approximately 0.89 overall accuracy but failed to detect alert cases, with 0% alert-class recall, showing that accuracy alone can be misleading in imbalanced monitoring scenarios. In the federated simulations, the model reached approximately 0.99 accuracy under clean labels, approximately 0.90 under 10% label noise, and approximately 0.70 under 30% label noise. Under a more difficult noisy, non-IID, dropout, and fixed skewed-client evaluation setting, the model stabilized at approximately 0.80 accuracy after 25 communication rounds. Client scaling from 5 to 20 clients remained stable, and FedAvg, weighted aggregation, and accuracy-trimmed robust aggregation produced similar final accuracy of approximately 0.98 in the 10-client setting. The results indicate that label quality strongly affects federated convergence, while blockchain anchoring should be interpreted as an auditability mechanism rather than a direct accuracy-improving component. This study provides a framework-level foundation for auditable federated patient monitoring in semi-trusted healthcare networks. Full article

EUPHEMIA, the Pan-European day-ahead electricity market-coupling algorithm, operates in a centralized manner that restricts independent auditability and has been characterized as pseudo-transparent. We propose a blockchain-based architecture that improves the transparency and verifiability of the market-coupling process while preserving participant confidentiality. It combines off-chain computation with selective on-chain publication and treats the three principal data categories of the EUPHEMIA pipeline separately: order books, network constraints, and clearing outputs. Zero-knowledge proofs utilizing zk-STARKs are employed to verify the integrity of the order book aggregation process and specific network-constraint sub-processes, whereas Merkle commitments ensure tamper-evident anchoring of publicly disclosed data. zk-STARKs are selected over CRS-based alternatives to eliminate the trusted-setup governance overhead associated with EUPHEMIA’s multi-jurisdictional structure. The estimated AIR trace size reaches approximately 2,225,000 rows in the worst-case NEMO (EPEX SPOT) scenario. A correction proof generated by the Regional Coordination Centre (RCC) requires an AIR of 641 trace rows when a binding network constraint exceeds its threshold during the review of Transmission System Operator (TSO) submissions. This trace size corresponds to an estimated proving time of 1 to 30 s based on reported STARK prover throughput. End-to-end verification of welfare maximization remains infeasible due to the lack of a complete public algorithm specification. Preliminary calibrated estimates are provided, and full empirical benchmarking remains as future work. Full article

Payroll-anchored retail borrowers—individuals whose monthly remuneration is routed into an account at the lending institution through a salary-project arrangement—constitute the volume backbone of unsecured consumer lending in Kazakhstan, generating the largest origination flow, the lowest realized default rate, and the majority of the systemic regulatory and capital sensitivities of second-tier banks. Payroll anchoring also changes the lender’s information set, which motivates a study of how that advantage translates into model performance and borrower outcomes. We design and internally validate an explainable hybrid artificial-intelligence framework stratified by client tenure into two production models: a Weight-of-Evidence (WOE) logistic-regression scorecard for new salary-project applicants, and a hybrid scorecard for repeat applicants, in which a stacked ensemble of LightGBM, CatBoost and a multi-head self-attention neural network contributes a single WOE-encoded predictor to a second-stage L2-regularized logistic regression. The hybrid recovers a substantial share of the ensemble’s discriminatory lift while preserving an auditable, monotone scorecard at the point of decision, and isotonic recalibration restores the predicted probabilities of default to the empirical bad-rate scale required for IFRS 9 expected-credit-loss accrual and risk-based pricing. We report discrimination, calibration and stability evidence under a strict anti-leakage protocol and set out the structural preconditions under which the architecture transfers to other emerging-market payroll-anchored portfolios. We are explicit about scope: a true out-of-time validation and a full group-conditional fairness audit are identified as required next steps rather than claimed here. The contribution is a reproducible, interpretable scoring design that exploits payroll visibility while retaining full coefficient interpretability inside the production decision engine. Full article

Real-time retail credit scoring is a data-intensive cognitive computing task. Each decision must fuse heterogeneous signals, execute a non-linear model, return a calibrated probability of default (PD), and emit a regulator-compliant local explanation within milliseconds. We address the most demanding segment of unsecured lending in Kazakhstan—Salary-Project-Independent (SPI) borrowers, whose principal income stream is not observable by the lender—and frame scoring as a constrained optimisation problem where we maximise discrimination subject to interpretability, latency, and calibration constraints. We propose a tenure-stratified hybrid framework that couples (i) an online weight-of-evidence logistic regression (WOE-LR) scorecard with (ii) an offline self-attention stacked ensemble (LightGBM, CatBoost, and a tabular self-attention network) whose calibrated PD is quantile-binned, WOE-encoded, and re-injected into the online scorecard as a single auditable predictor. On 551,962 production contracts that originated in 2022–2024, the repeat-client hybrid attains an area under the receiver operating characteristic curve (AUROC) of 0.826, a Gini coefficient of 0.65, and a Kolmogorov–Smirnov (KS) statistic of 0.495, preserving roughly half of the offline ensemble’s lift over the linear baseline (AUROC 0.79→0.897) while retaining a fully auditable twelve-coefficient scorecard in production. The new-client scorecard attains an AUROC of 0.741. Non-parametric isotonic recalibration reduces the expected calibration error from 0.27 to below 0.01 and raises the Hosmer–Lemeshow p-value above 0.99 without altering discrimination. The framework complies with the model risk standards of the Agency of the Republic of Kazakhstan for Regulation and Development of the Financial Market and is delivered as a Spark/MLOps reference architecture, illustrating how big data engineering, attention-based representation learning, and post hoc explanations can be co-designed for a high-stakes, high-throughput, regulated AI application. Full article

High photovoltaic (PV) penetration supports low-carbon distribution networks, but reverse power flow can drive radial feeders beyond voltage limits and reduce hosting capacity. Local droop control avoids communication but has limited coordination capability, whereas centralized regulation relies on global measurements and repeated computation. This paper proposes a nonlinear feasibility-recovery distributed online primal–dual Push–Sum framework (NFR-DOPP) for coordinated PV-inverter reactive-power control. The framework combines projected primal–dual updates with sparse directed Push–Sum coordination and a nonlinear recovery term that strengthens correction when voltage constraints become active. An error-compensated Top-k differential compression layer further reduces inter-agent state exchange while preserving the physical feedback loop and projected update. Nonlinear closed-loop simulations on a modified IEEE 123-bus feeder show that NFR-DOPP restores voltage feasibility more effectively than conventional distributed and local droop baselines. Under a 06:00–18:00 dynamic profile, the compressed implementation maintains zero PV node voltage violations and reduces cumulative transmitted bits by approximately 31% relative to full-state communication. A secondary active-loss audit in the tested case indicates that stronger coordinated reactive-power regulation may increase losses. The proposed framework should therefore be viewed as a voltage-feasibility and communication-efficiency method rather than a loss-minimization strategy. Full article

This Perspective examines a central problem in food safety governance: why organisations with well-developed food safety management systems, standard operating procedures, and audit mechanisms may still struggle to achieve stable and consistent frontline food safety practices. Using a narrative integrative review approach, this paper draws on the literature on food safety culture, organisational behaviour, institutional theory, implementation Champions, and frontline practice. It argues that the food safety is not only dependent on the effectiveness of the formal food safety systems, but also on how institutional expectations are interpreted, negotiated, reinforced, and enacted in everyday work. This paper introduces the concept of Food Safety Champions and conceptualises them as translational actors who can help connect formal food safety requirements with situated practice through meaning making, contextual negotiation and behavioural reinforcement. It further suggests that a Champion’s effectiveness depends not only on individual initiative, but also on peer trust, organisational legitimacy, their informal networks and connections, enabling conditions, and continuous institutional feedback. This paper contributes a practice-based conceptual framework for understanding how formal food safety systems may become more meaningful, sustainable, and behaviourally effective in food production environments. Full article

Phishing email detection is a practical requirement for human-operated security and administration workflows, including mail gateways, security operations centre (SOC) triage queues, cloud dashboards, firmware-management portals, and other operational interfaces. In Internet-of-Things (IoT) and wireless sensor network (WSN) operations, compromise of such human-operated interfaces can open an initial-access path into the sensor-network management plane; however, the present study evaluates email body classifiers on public phishing email corpora rather than on-device IoT/WSN hardware. Email filtering must often run on low-cost servers, where running and storing a transformer for every message is costly in compute and memory, and opaque scores are hard to audit. We compare four model families on four public phishing email corpora: hand-engineered classifiers, high-vocabulary classical models, frozen-transformer classifiers, and a CPU-feasible fine-tuned DistilBERT reference. The strongest low-cost model is a Platt-calibrated Linear SVM with high-vocabulary TF-IDF features. It reaches F1 = 0.961 on the primary corpus, slightly exceeds the frozen DistilBERT baseline and is statistically indistinguishable from the fine-tuned DistilBERT reference in-domain, and requires 0.23 MB on disk with about 8 ms per email. This shows that transformer inference is not required for competitive accuracy on the evaluated in-domain corpora. We then evaluate the Calibrated Selective Cascade (CSC) as an operational routing layer: a calibrated low-cost arm handles most messages, while a validation-selected uncertainty band is deferred to a transformer arm. With the strong SVM arm, CSC yields only marginal in-domain F1 gains, but provides a tunable latency/deferral trade-off parameter and a monitorable drift signal. On an exploratory synthetic stress set of stylistically neutral, LLM-style phishing, no evaluated body-only model exceeds F1 around 0.46; closing this gap would require header-, URL-, identity-, and attachment-level signals. Full article

Byzantine-resilient multi-agent reinforcement learning (MARL) matters in networked cyber-physical systems, where corrupted sensor messages degrade formation accuracy and execution-time safety. This paper presents an evaluation and audit study: a multiplicity-corrected operating-regime protocol applied to RS-MARL, a representative trust-based safety pipeline. The aim is to identify supported, inconclusive, and detector-limited regimes rather than claim a universally superior new MARL algorithm. The evidence base contains a 3000-run core matrix over five methods, six attack families, five Byzantine ratios, and 20 seeds per cell; 580 benign-control and ablation runs; and a 2380-run review-audit extension covering A-CBF calibration, four-switch ablation, sensor impairment, and high-seed confirmation. Results are regime-specific. RS-MARL has lower mean safety violations than Safe-MAPPO in 19 of 30 attack-ratio cells, but no core contrast survives Holm correction. Detection is reliable under collusive, random, and stealthy attacks, but weak or undefined under constant, adaptive, and sign-flip attacks, which bound the current energy-based trust detector’s operating envelope. A-CBF margin retuning does not improve over the deployed setting after correction, while four-switch ablation identifies SET as independently necessary for collusive-attack detection. The results support a reproducible reporting template: matched baselines, sensitivity estimates, detection reliability, artefact audits, and explicit safety-performance trade-offs. Full article