Search Results (100)

LLM privacy risks arise across different lifecycle stages and architectural boundaries, and existing protection mechanisms provide only partial coverage. This paper analyzes the main families of privacy-preserving approaches for LLM systems through a two-axis structure that crosses lifecycle stages with system architecture layers. Some safeguards are operationally mature; others, such as confidential computing, have moved into production practice; stronger cryptographic methods, while most promising in principle, remain research-heavy in practice. No single mechanism provides complete end-to-end protection: different methods protect different assets, operate at different lifecycle stages, span distinct system layers, and carry distinct trust, performance, and deployment trade-offs. Practical LLM privacy is therefore a problem of layered system design rather than the search for a universal primitive, and hybrid architectures are emerging as the most realistic deployable pattern. Building on this analysis, we propose a six-dimensional evaluation framework for privacy in hybrid LLM deployments (a 0–5 ordinal scoring rubric designed for reproducible application, with explicit anchor language and per-score evidence requirements) and apply it to five representative confidential AI deployments, deriving the scores in full for two of them. The framework feeds a three-tier gap-closure roadmap and design principles for architecture-time use, connecting what privacy technologies promise, what they actually protect, and what is realistically deployable in modern LLM systems. Full article

Soil salinity is a major constraint on global wheat productivity, yet the genetic and molecular determinants of root system architecture (RSA) adaptation under salt stress remain poorly characterized. We integrated a genome-wide association study (GWAS) of 566 wheat accessions with comparative RNA-seq transcriptomics to identify the genetic and transcriptional determinants of RSA adaptation under 200 mM NaCl. GWAS identified a candidate locus on chromosome 7B harboring TaIAO, which encodes a protein with predicted aldehyde oxidase-like activity consistent with a role in tryptophan-dependent auxin biosynthesis. Accessions carrying the favorable CC allele exhibited significantly greater root volume retention than those carrying the GG genotype (p < 0.001). Comparative RNA-seq revealed that the salt-tolerant Sarajevo 1 exhibited coordinated transcriptional repression of three distinct modules—cell wall expansion (TaExpansin), auxin redistribution (TaPIN-like), and stress-associated ROS defense (TaPOD1)—whereas the sensitive genotype CI 17260 aberrantly induced or incompletely repressed these modules under stress. ELISA-based IAA quantification, ROS imaging, and qRT-PCR analysis provided independent physiological and transcriptional support for these patterns. These findings support a two-tier adaptive model in which constitutive genetic variation at the TaIAO locus may contribute to a developmental baseline, coupled with coordinated stress-responsive transcriptional repression of energy-consuming modules, providing promising targets for marker-assisted breeding of salt-tolerant wheat. Full article

Point-of-sale (POS) consumer credit represents the most rapidly expanding retail-lending channel within the emerging Eurasian markets, necessitating a stringent operational framework for the underwriting model: the decision must be rendered within a mere few hundred milliseconds during the in-store checkout process, while the inputs are constrained to what the application XML is capable of conveying. This research endeavors to develop, internally validate, and operationally delineate a hybrid, explainable artificial intelligence framework aimed at POS credit scoring within the production portfolio of Kazakhstan’s largest second-tier bank. The architectural framework is delineated along two orthogonal dimensions—client tenure and decision-making channel—resulting in the formulation of three distinct production models: two transparent Weight of Evidence–Logistic Regression scorecards tailored for the real-time channel, and one isotonically-calibrated stacked ensemble (comprising LightGBM, CatBoost, and a three-layer neural network) designated for the batch channel. The selection of hyperparameters was conducted utilising Bayesian optimization within the context of stratified five-fold cross-validation. The digital scorecards achieve an area under the receiver operating characteristic curve (AUROC) of 0.847 and 0.835, whereas the offline ensemble enhances performance to an AUROC of 0.918, accompanied by a Kolmogorov–Smirnov statistic of 0.682 and a Gini coefficient of 0.836. The population stability indices persist below the threshold of 0.07, while isotonic recalibration effectively reduces the Brier score by 18%. Furthermore, an extensive examination of fairness demonstrates variations in approval rates within a margin of ±1.2 percentage points—and equalised-odds gaps below 1.5 percentage points in the true-positive rate and 0.7 percentage points in the false-positive rate—across multiple demographic factors such as gender, age, and distinctions between urban and rural classifications, thus establishing an artificial intelligence framework that is both regulatorily compliant and interpretable, aligning with the directives set forth by the Agency of the Republic of Kazakhstan for Regulation and Development of the Financial Market. Full article

The rapid advancement of quantum computing poses a fundamental threat to classical public-key cryptographic systems, necessitating the transition to post-quantum cryptography (PQC). While significant progress has been made in the standardization of quantum-resistant algorithms, their practical deployment in heterogeneous environments—particularly resource-constrained Internet of Things (IoT) devices—remains a critical challenge. This study presents a comprehensive experimental evaluation of four NIST-standardized PQC algorithms: CRYSTALS-Kyber (ML-KEM), CRYSTALS-Dilithium (ML-DSA), FALCON, and SPHINCS+. The scope of these findings is bounded by an empirical analysis conducted across two specific testing platforms, a high-performance x86-64 workstation (AMD Ryzen 7 5700U) and a resource-constrained embedded microcontroller (ESP32-WROOM), utilizing dedicated software environments implemented in Native C, Go, and Python. The evaluation isolates key performance indicators, including computational latency, memory consumption, communication overhead, and temporal determinism, based on benchmarking over 1000 iterations. Within this experimental setup, results demonstrate clear trade-offs between target security categories, execution performance, and structural memory limits. Lattice-based schemes such as Kyber and Falcon exhibit optimal efficiency and scalability on the tested embedded platform, while the specific memory limits of the ESP32 platform introduce architectural stability constraints for higher-tier Dilithium variants. In contrast, SPHINCS+ provides structural robustness at the cost of higher computational hashing latency within these evaluation environments. The findings highlight the critical role of hardware-specific constraints and language runtime design choices in enabling practical PQC deployment, providing context-specific insights supporting the secure migration of IoT infrastructures toward quantum-resilient systems. Full article

Cyanobacteria inhabit ecosystems ranging from oligotrophic deserts to eutrophic lakes, yet it remains unclear whether distantly related species dominate in disparate habitats, share common genomic features, or show divergent specialization. Here, we established a comparative framework of Microcoleus vaginatus, the pioneer stabilizer of biocrusts, and Microcystis aeruginosa, a major cause of freshwater blooms worldwide. Our dataset comprises 504 high-quality cyanobacterial genomes, including 132 M. vaginatus, 148 M. aeruginosa, and 224 reference taxa, for analyses of genome architecture, functional repertoires, and genomic plasticity. Both focal lineages showed signatures of extensive horizontal gene transfer and shared a small set of conserved orthologous groups, annotated as FAD-dependent oxidoreductases, manganese efflux, and class II aldolases. Nevertheless, the two lineages followed distinct genomic strategies. M. vaginatus expands regulatory breadth and stress-resilience gene families, whereas M. aeruginosa shows evidence of genome streamlining and rapid nutrient exploitation. Notably, we hypothesize that aquatic M. vaginatus strains retain ancestral terrestrial genomic features while gradually acquiring potential aquatic-specific adaptations. Together, these results reveal a two-tier architecture associated with cyanobacterial dominance and provide a testable hypothesis for how cyanobacterial lineages may respond to global change pressures. Full article

Climate-related disasters are escalating in frequency and severity, yet global adaptation finance remains critically insufficiently structured to respond after disasters occur rather than before. This study empirically examines disaster loss data, climate finance flows, and financial instrument evidence to test two hypotheses: whether climate finance is disaster-reactive, and whether policy uncertainty constrains it. We integrate data from the Emergency Events Database (EM-DAT), covering seven climate-induced hazard types (droughts, extreme temperatures, floods, glacial lake outburst floods, wet mass movements, storms, and wildfires), in addition to the OECD Creditor Reporting System (CRS), the World Uncertainty Index (WUI), the ND-GAIN vulnerability index, and the World Governance Indicators, the Green Climate Fund Open Data Library, and the Artemis Deal Directory across 131 countries (2011–2024) for Hypothesis 1 and 100 countries (2012–2024) for Hypothesis 2. Fixed-effects panel regressions with Driscoll–Kraay standard errors confirm that prior-year disaster losses significantly predict subsequent climate finance flows (β = 0.040, p = 0.009; N = 1769 country-year observations), establishing a reactive financing pattern. Policy uncertainty interacting with high vulnerability is found to suppress adaptation finance flows (β = −2.587, p = 0.080, N = 878 country-year observations), with the effect concentrated among the most climate-exposed economies. We propose a risk-layered climate finance architecture aligning instruments with distinct hazard tiers across the Global South. Credible policy signals, strategic public investment, and systematic integration of insurance mechanisms are essential preconditions for unlocking scalable, forward-looking resilience finance. Full article

Artificial Intelligence of Things (AIoT) is transforming Internet of Things (IoT) systems from cloud-centric data processing into distributed intelligence across device, edge, fog, and cloud tiers. However, existing reviews often emphasize specific computational layers, learning paradigms, or application domains rather than the cross-domain problem of Artificial Intelligence (AI) workload placement under real deployment constraints. This paper presents a structured integrative review of AI workload placement in AIoT, based on a multi-stage literature search, two-stage screening process, and thematic synthesis of 132 sources. The review does not propose a new physical architecture; instead, it develops a terminology-harmonized and AI-centric perspective for assessing where AI functions should reside according to latency, privacy, bandwidth, power, scalability, resilience, and model complexity. Evidence is synthesized across Industrial Internet of Things (IIoT), smart cities, Internet of Medical Things (IoMT), and smart agriculture. The findings show that placement drivers are domain-dependent: deterministic response and reliability dominate IIoT, interoperability and scale shape smart cities, privacy and human oversight constrain IoMT, and energy scarcity and intermittent connectivity define agriculture. The review concludes that robust AIoT requires hybrid multi-layer architectures combining Tiny Machine Learning (TinyML), edge/fog coordination, cloud-scale optimization, and Federated Learning (FL) where appropriate. Full article

Conventional multi-criteria decision-making approaches for battery energy storage system (BESS) dispatch evaluation treat regulatory and policy conditions as compensable criteria within a single aggregate score. This becomes problematic when institutional admissibility functions as a prerequisite for deployment rather than a tradeable attribute. This study aims to develop and test a sequential gate framework. The methodological contribution lies in the evaluation architecture itself: the framework distinguishes sequential admissibility gating from conventional compensatory Multi-Criteria Decision-Making (MCDM). Deployment feasibility is conceptualized as a layered construct in which regulatory admissibility defines the feasible solution space and technical performance differentiates among admissible options. The framework integrates systematic literature screening, quantitative policy and regulatory assessment, and technical ranking using a hybrid Best-Worst Method, Entropy weighting, and TOPSIS approach. A Danish case study covering twelve dispatch strategies compares the proposed sequential design with two flat alternatives. The results show that the evaluation architecture materially affects outcomes: sequential gating excludes an institutionally incomplete strategy and reorders the upper tier by removing compensatory policy effects. Coordinated multi-BESS control at Electric Vehicle charging parks achieves the highest combined feasibility (closeness coefficient 0.891, ranked 1st), while mobile BESS is excluded by the admissibility gate. The sequential design reorders the upper tier relative to flat MCDM, with S4 and S6 rising and S2 and S10 falling once policy compensation is neutralized after the gate. The top-ranked strategy remains robust across sensitivity analysis, Monte Carlo simulation, score perturbation, and VIKOR cross-validation. The framework is presented as an analytical pre-simulation screening tool rather than a validated implementation instrument; external validation against real deployment outcomes is identified as a priority for future research. The framework provides a structured, decision-consistent approach for evaluating deployment feasibility in regulated energy systems. Full article

Adaptive and evolutive software systems are characterized by ontologically defined non-determinism—not a defect but the primary force of their evolution. Non-determinism arises from recursion, interaction, and selection between abstract components and can only be resolved as the execution sequence grows sufficiently for one outcome to become determinate. In adaptive systems, managing this non-determinism through structural adaptation of abstract components during execution is the defining operational characteristic—one that no existing execution substrate formally supports. The problems of evolutive AI systems—inconsistency, non-reproducibility, absence of causal traceability, and an inability to enforce purpose-constrained autonomy—cannot be resolved within AI architectures alone. Resolving them requires a formal execution substrate in which causal context growth, resolution-moment detection, and structural adaptation of abstract components are first-class properties. This paper introduces the Zero Tier Execution Substrate (ZTES), grounded in a foundational model that defines execution as a sequence generated by recursive invocations of abstract components with ontologically specified purpose, in which non-determinism is resolved within the causal context of the sequence before commitment. ZTES is a homomorphic specification of this model, achieved through disciplined composition of the Mesarović–Takahara system ontology, Lamport-consistent causal ordering, P-DEVS transition semantics, and the Three-Phase execution kernel—mechanisms individually proven at a global scale. System execution is formally identified with the causal evolution of knowledge: Execution(Σ) ≡ Evolution(K). The historical knowledge base K has a two-dimensional orthogonal structure—the eschatological dimension encoding purpose lineage through recursive specialization, and the sequential dimension encoding event order through iterative mapping—which establishes purpose integrity as a substrate-level property of evolutive execution. The semantic closure of ZTES establishes deterministic reproducibility, governance–execution equivalence, and purpose-constrained autonomy as structural consequences of substrate closure rather than as additional architectural layers. Full article

Australian local governments are increasingly deploying smart city technologies to manage climate-related shocks and chronic stresses, yet implementation often remains fragmented and difficult to embed in routine practice. Many initiatives stall in “pilot-forever” cycles because decision rights, equity safeguards, operational integration, and learning systems are applied inconsistently. This paper introduces the Smart Urban Resilience Framework (SURF), a phase-gated, tier-aware governance framework designed to support the institutionalisation of smart urban resilience through more transparent and evidence-based decision-making. The SURF is grounded in an integrated evidence-to-design synthesis drawing on a systematic review, a comparative analysis of Tier 1 and Tier 2 Australian local government strategies, an in-depth Sydney case study, and stakeholder interviews. Although empirically grounded in Australian local government, the SURF is designed as a governance architecture that may be adapted in comparable municipal settings elsewhere. The framework comprises a staged pathway, two evidence gates, and four concurrent action tracks, supported by enabling layers and traceable evidence tools. The SURF is presented as a practical implementation architecture intended to support more transparent and defensible decisions about funding, scaling, refining, or retiring smart resilience initiatives. In this paper, resilience is operationalised through a service continuity lens, focusing on how digital initiatives can be embedded in governance and delivery systems to support the continuity of essential local government services under stress. Full article

The space–air–ground integrated network (SAGIN) is a key 6G architecture that provides seamless three-dimensional connectivity, exhibiting hierarchical structural symmetry between LEO satellite and HAP layers. Integrating information-centric networking (ICN) with caching on Low Earth Orbit (LEO) satellites and high-altitude platforms (HAPs) significantly enhances content distribution efficiency. Existing studies on caching mechanisms have made progress but lack optimized cache resource allocation and accurate popular content identification. Thus, an interest-aware caching scheme (ICRL) based on reinforcement learning is proposed to optimize the SAGIN’s popular content caching decisions, aiming to achieve rational symmetric allocation of cache resources across LEO and HAP layers. Different from existing RL-based caching methods, the proposed ICRL scheme considers the LEO-HAP hierarchical architecture and designs an improved reinforcement learning mechanism to adapt to the dynamic characteristics of the SAGIN. First, an air–space two-tier caching architecture is constructed to enable collaborative caching between LEO satellites and HAPs. Second, to select high-value nodes intelligently, the proposed scheme leverages a comprehensive importance model that quantitatively analyzes HAP and LEO indicators such as topology, transmission capacity, and location. Finally, a reinforcement learning-based dynamic cache mechanism is developed. It captures real-time network requests and cache states to select optimal actions and adapt to network dynamics for better content popularity matching. Extensive evaluations based on NDNSIM demonstrate that ICRL outperforms baseline schemes in terms of cache hit ratio, server load, and request latency and achieves a symmetric balance of network load and service performance in the whole SAGIN. Full article

Digital learning tools for American Sign Language (ASL) often lack the interactive depth necessary to engage learners effectively. This paper introduces a novel, browser-based word search game designed to facilitate ASL vocabulary familiarization through gamified interaction. The system employs a two-tier architecture consisting of a React-based frontend and a Flask-based backend. At its core, the application integrates a lightweight, skeleton-based Isolated Sign Language Recognition (ISLR) model, utilizing a Stacked Transformer-based Spatial-Temporal Attention Network to enable real-time webcam-based word entry during the configuration phase. This model, trained on the WLASL-100 dataset, achieves a Top-5 test accuracy of 88.48% with an average model inference latency of 141 ms, enabling real-time webcam input without proprietary hardware. Furthermore, we implement a constraint-satisfaction puzzle generation algorithm that achieves a 100% success rate in creating interlocked, multi-directional grids. Our results demonstrate that merging computer vision with pedagogical game mechanics provides an accessible, high-performance tool for the Deaf and Hard-of-Hearing (DHH) community, bridging the gap between static instruction and active linguistic practice. Full article

The rapid proliferation of artificial intelligence (AI) has pushed hyperscale data center rack densities beyond 100 kW, driving facility power requirements to the gigawatt scale. As developers attempt to deploy these massive Zettascale compute loads across US wholesale electricity markets, they encounter severe transmission planning bottlenecks, multi-year interconnection delays, and escalating grid transient stability risks. This paper presents a generalizable techno-economic framework for evaluating grid-tied, behind-the-meter (BTM) energy architectures as a means of bypassing these constraints. The framework is demonstrated through a detailed case study in the Electric Reliability Council of Texas (ERCOT), selected for its rapid data center growth and evolving large-load regulatory environment. Using a scenario-based comparative approach, this study models the feasibility of transitioning from pure-grid reliance to hybrid, on-site generation across a three-phase deployment pathway scaling from 25 MW to 250 MW. Six distinct microgrid configurations are evaluated, integrating baseload technologies—including Enhanced Geothermal Systems (EGSs), Small Modular Reactors (SMRs), and Reciprocating Internal Combustion Engines (RICEs)—with a tiered-performance Battery Energy Storage System (BESS) combining high C-rate lithium-ion units and repurposed electric vehicle batteries. System viability is assessed through two primary metrics: the Levelized Cost of Energy (LCOE) and the Avoided Loss of Load Probability (ALOLP). The results indicate that the blended LCOE scenario ranges from $64.50/MWh (Geothermal + Solar PPA) to $94.20/MWh (SMR-anchored), compared to a $75.00/MWh pure-grid baseline. The 100% Geothermal configuration achieves a scenario-dependent ALOLP exceeding 99.9%, while gas-dependent configurations range from 58.0% to 91.2%. These findings suggest that geographic siting co-optimized with localized generation offers a viable pathway for balancing regulatory compliance, capital cost, and Uptime Tier IV operational resilience in early-stage data center development across constrained grid environments. Full article

This paper proposes a CBDC design trilemma, the claim that central banks cannot simultaneously maximize privacy, financial stability, and regulatory compliance when designing retail central bank digital currencies and finds the existing literature consistent with this proposition. Through a systematic review of 140 peer-reviewed articles (Web of Science SCIE/SSCI indexes, 2014–2026, supplemented by Scopus and SSRN), evidence is synthesized across four thematic dimensions: design frameworks and architecture, financial stability and banking risk, privacy and security trade-offs, and user adoption and institutional quality. Cross-tabulation of coded data supports all three pairwise tensions: privacy-enhancing designs weaken AML/CFT enforcement, anonymous holdings amplify bank-run risk, and stringent prudential safeguards constrain transaction monitoring. The literature converges on two-tier, hybrid architectures with tiered privacy as the dominant compromise a “zone of feasible design”, that sacrifices full optimality on each vertex. Nine research gaps are identified, most critically the scarcity of empirical evidence from live deployments, the neglect of wholesale CBDC, and insufficient analysis of cross-border interoperability. The framework offers policymakers a structured lens for evaluating retail CBDC design trade-offs and researchers a testable proposition for future empirical work. Full article

Background: Explainable artificial intelligence (XAI) is increasingly proposed to improve trust in mammography-based artificial-intelligence systems, but the validity and clinical readiness of published explanations remain unclear. We aim to systematically review XAI methods applied to mammography and synthesize how explanations are evaluated for validity, robustness, and clinical usefulness. Methods: We conducted a systematic review according to PRISMA 2020. MEDLINE/PubMed, Embase, Scopus, Web of Science Core Collection, and the Cochrane Library were searched from 1 January 2015 to 15 January 2026. Two reviewers independently screened records and extracted data; disagreements were resolved by consensus with a third reviewer. Included studies used mammography as the primary input and reported an explicit explanation or interpretability mechanism. Because the literature was methodologically heterogeneous, we performed a structured narrative synthesis and an adapted XAI-specific appraisal of explanation claims, quantitative evaluation, external validation, human-factor assessment, and reporting transparency. Results: Fourteen studies were included. Ten studies addressed detection or lesion classification and four addressed risk or outcome prediction. Primary XAI families were interpretable-by-design architectures (6/14), post hoc saliency or attribution methods (5/14), and feature-level explanation methods (3/14). Five studies remained at tier-1 qualitative plausibility only, seven reached tier-2 internal quantitative explanation evaluation, two reached tier-3 external or cross-dataset interpretability assessment, and none reported reader or workflow studies. In the dedicated mammography saliency benchmark, Pointing Game scores for Grad-CAM, Grad-CAM++, and Eigen-CAM ranged from 0.30 to 0.41, indicating only modest lesion-pointing reliability despite acceptable classifier performance. Conclusions: Mammography XAI remains dominated by visually plausible explanations that are inconsistently validated. The literature is moving toward task-aligned and intrinsically interpretable designs, yet external validation and clinician-centered evaluation remain rare. Future studies should pre-specify explanation claims, use task-appropriate quantitative metrics, report explanation robustness under distribution shift, and test whether explanations improve human decision-making. Full article