Search Results (451)

Recursive evidence fusion gives an intrusion detection system temporal memory, but it also gives unreliable windows and erroneous review outcomes a path to influence later diagnoses. Existing drift-handling, open-set, conformal, continual-learning, and human-in-the-loop methods provide useful signals or update classifiers and memories; they do not, by themselves, specify when a post-classification evidential state may be written or replaced. We present RTEF-IDS, a protocol that separates current action, model-evidence admission, reviewed-feedback admission, and history replacement. The protocol retains the history-relative reliability principle from our previous work, instantiates it for singleton-plus-ignorance IDS evidence, and assigns operation-specific credentials. Reviewed windows make no base-state change, mapped-known feedback may be appended, and replacement requires persistent confirmation. On 33,384 frozen windows, 30% retrospective admission excludes 26.3% of held-out-or-misclassified mass while retaining 94.8% of known-correct evidence. Under paired imperfect feedback, retrospective replacement increases one-window future history-state agreement by 0.107 in the primary block and 0.129 in IoT-23 leave-scenario-out replay. Under a past-only rolling-budget gate within externally supplied frozen partitions, the corresponding increments are 0.001 and 0.000, indicating that the tested gate exposes few qualifying replacement opportunities; bounded external short streams show the same opportunity constraint. Independent second review reduces false authorization from 5.66 to 0.124 per 1000 first-stage reviewed windows under independent errors and from 34.27 to 0.181 under five-window correlated errors, with a corresponding increase in review demand and a reduction in admitted corrective feedback. A shared systematic label alias remains unresolved by the tested review arms. These results support explicit, auditable state-mutation control while identifying the causal-opportunity and feedback-provenance conditions under which it operates. Full article

We present ESNformer, a hybrid architecture that couples an Echo State Network (ESN) reservoir with a Transformer encoder for classification of structured, multi-indicator assessment data: a fixed-order vector of complementary indicators per assessment instance rather than a repeated-measures time series. The reservoir acts as a fixed nonlinear feature map over the indicator vector, while self-attention, made position-aware over the fixed column order, learns how each indicator’s evidence contributes to the final decision, so the two components, together, capture local, indicator-level detail and global, cross-indicator interactions within a single, end-to-end trainable model. Interpretability is treated as a first-class design requirement rather than an afterthought: the architecture is paired with an explainability layer combining SHAP feature attribution (reported both globally and per class), the model’s own attention weights, a deletion/insertion faithfulness test that quantitatively verifies which inputs the model actually relies on, and counterfactual maps that translate a prediction into an actionable, inspectable recommendation. We evaluate the architecture on a concrete case study, classifying Braille-literacy instructional recommendations from 15 pedagogical indicators grouped into three categories (Mangold’s, ABKL, and Progresar), using a benchmark of 900 real assessment instances (630 used, together with a class-conditional augmentation procedure, to build a 2100-instance training set) with validation and test partitions (135 instances each) kept exclusively real. On this benchmark, the tuned model reached 85.33% accuracy, 85.90% macro-precision, 85.33% macro-recall, an F1 score of 85.25%, and an AUC of 0.95 on the real test set. SHAP attribution, attention weights, and the faithfulness test converge on the same two dominant indicators (response time and error count): removing them alone collapses accuracy to chance, while retaining only them recovers most of the model’s accuracy. We report this transparently alongside a comparison against ESN-only, Transformer-only, and tabular baselines (logistic regression, decision tree, random forest, XGBoost, and an MLP) on the same data and discuss what the hybrid architecture and its explainability pipeline add beyond what the two dominant indicators already explain and how the approach generalizes to other tabular and mixed-granularity assessment settings that require both predictive accuracy and a verifiable account of what drove each decision. Full article

Inland container depot (ICD) truck schedules must absorb new orders, service delays, appointment changes, congestion, and port cut-offs without destabilizing an already executed plan. This study asks whether event-triggered local repair can be separated into an explicit business-rule audit and a learned ranking of feasible task–vehicle actions. The proposed decision-support framework connects a static baseline, candidate task chains, six modeled hard-feasibility predicates, a Transformer encoder trained with proximal policy optimization (Transformer-PPO), and discrete-event execution logs. A five-seed, 120-episode confirmation gave Transformer-PPO a held-out online completion proxy (α_online) of 0.3226 and reward of 110.58, compared with 0.2581 and 61.87 for the matched multilayer perceptron (MLP); deterministic rules and search remained competitive. An independent audit of 4,968,000 action cells across 552 decision states found no disagreement with an independently coded oracle for the implemented hard predicates, while a reward-weight screen exposed the expected efficiency-stability trade-off. Together with a rolling-horizon comparator and a three-scale by three-disturbance stress test, the evidence supports an auditable system-integration contribution, not a new generic reinforcement learning (RL) algorithm or universal performance superiority. Claims are limited to synthetic simulation-based decision support. Full article

Self-determination is critical in promoting classroom engagement among students with autism spectrum disorder. The Self-Determined Learning Model of Instruction (SDLMI), an evidence-based practice for enhancing self-determination in students with disabilities, has gained increasing scholarly attention. However, empirical studies in Chinese contexts remain limited. This exploratory, descriptive implementation study used an A–B–M single-case approach with two individual cases to examine descriptive patterns in self-determination skills, perceived opportunities for self-determination, and classroom engagement before, during, and after the implementation of an SDLMI-based action-plan in Mandarin-speaking special education classrooms in Fujian Province, China. Preliminary findings indicated positive trends in students’ self-determination skills and classroom engagement after the implementation of the SDLMI, as well as increased perceived school-based opportunities for self-determination and mixed changes in perceived home-based opportunities. These findings are consistent with the existing international literature and provide preliminary case-level support for the cultural feasibility of adapting SDLMI practices within Chinese special education settings. Further research with larger samples and more rigorous experimental designs must fully assess its effectiveness and generalizability. Full article

Background/Objectives: We aimed to evaluate the methodological quality, translational limitations, and clinical applicability of current artificial intelligence (AI) applications in periodontology and to propose a framework for validated prediction and decision support. Methods: A structured narrative review based on a targeted, non-systematic literature search was conducted using PubMed and cross-disciplinary sources (January 2015–April 2026). Evidence from primary studies, systematic reviews, and methodological guidance for AI prediction models and clinical decision-support systems was synthesized with a focus on clinical applicability. Results: Current periodontal AI research is dominated by retrospective studies focusing on radiographic phenotyping, where deep learning models demonstrate promising diagnostic performance for detecting and quantifying periodontal bone loss. However, substantial limitations persist, including heterogeneous endpoints, inconsistent reporting, limited external validation, and insufficient calibration assessment. Importantly, there is little evidence that AI-based tools improve clinical decision-making or patient-relevant outcomes. Emerging work on prognostic modeling and multimodal data integration highlights the potential for individualized periodontal risk prediction but remains undervalidated, with limited evidence for clinical implementation. Conclusions: Although AI-based models show promising diagnostic performance, translational progress in periodontology is currently limited by insufficient validation and the lack of evidence for clinical utility. Future research should prioritize clinically actionable prediction models, robust external validation, and prospective evaluation of AI-supported decision-making within real-world periodontal care pathways. Full article

After infrastructure disruption, an emergency wireless controller must meet urgent communication demand and preserve resources for later periods. We propose Scarcity-Coefficient Gated Projection Reinforcement Learning (SCGP-RL). It jointly selects total planning-layer activation and a regional capacity upper-bound vector. Scarcity and urgent-demand evidence shape the activation intent. Scalar and capped-simplex projections enforce the coupled action constraints. A planning capacity unit (PCU) is defined as a calibratable service-capacity quantum. In the common constrained evaluation, SCGP-RL reduced the unmet urgent-demand score from 0.6643 for Projected CPO to 0.4919. It also satisfied all the executed hard constraints. Component tests show that the urgent-demand gate drives rapid service response and that the marginal demand-relief estimate provides a smaller benefit. Binding-condition tests show that the power, backhaul, and node-health mechanisms protect the resources they represent. Full article

Market gardening is increasingly discussed as a sustainability-oriented farming model with the potential to support resilient, locally embedded food systems and to contribute to food security through diversified fresh-produce supply, short value chains, and closer producer–consumer relations. However, limited empirical evidence exists on the competences and training needs required for its professional and agribusiness development. This study examines perceived competence confidence and training priorities for market gardening across Ireland, Greece, France, Germany, and Austria. The research draws on 53 interviews with education and training providers and a cross-sectional survey of 275 respondents, including practitioners, advisory actors, and institutional stakeholders. The survey was structured around EntreComp, DigComp, and GreenComp to assess entrepreneurial, digital, and sustainability competences. The findings show that practitioners reported relatively strong confidence in motivation, vision, ethical thinking, environmental responsibility, and local contextual awareness, but weaker confidence in market research, resource mobilisation, record-keeping, planning, digital farm management, online sales, technical troubleshooting, stakeholder engagement, and participatory sustainability action. Advisory and institutional respondents rated entrepreneurial and sustainability competences as highly important, while digital competences were also considered relevant but showed stronger cross-country variation. The study concludes that market gardening education should move beyond technical horticultural instruction and adopt an integrated, competence-based curriculum combining entrepreneurial, digital, green, and practice-oriented learning. The findings highlight competence-development priorities that may strengthen market gardening and its potential contribution to local food availability, resilient short value chains, and food security. Full article

The DC–DC boost converter is a challenging control target because of its nonlinear dynamics, wide operating range, and non-minimum-phase behavior under continuous conduction mode. These control challenges are particularly pronounced under large-signal transients, parameter variations, constant-power-load effects, and hardware constraints. This review examines deep reinforcement learning-based control of DC–DC boost converters from an engineering-oriented perspective. It covers learning-assisted classical control, direct duty-cycle control, and hybrid architectures, with attention to action design, reward formulation, observation timing, safety constraints, and validation fidelity. A structured search of Scopus, Web of Science Core Collection, and IEEE Xplore was used to identify boost-specific studies and transferable adjacent-converter evidence. Rather than ranking algorithms alone, the review organizes the literature around converter-aware and hardware-oriented learning control. The review argues that recent progress should not be interpreted as a simple replacement of classical control by deep reinforcement learning. Accordingly, algorithm choice, physical knowledge, action and reward design, observation timing, safety constraints, and validation fidelity are treated jointly. The available evidence suggests that progress toward credible practical deployment requires integrating converter physics, bounded or hybrid control authority, explicit safety constraints, and hardware-oriented validation. Full article

Reinforcement learning (RL) has emerged as a promising approach for autonomous surgical robotic subtasks. Recent advances include deep reinforcement learning (DRL), imitation learning (IL), and vision–language–action (VLA) models. However, current evidence remains fragmented across simulation benchmarks, task-specific demonstrations, and limited clinical studies. Existing reviews primarily focus on RL algorithms, while the broader pathway from algorithm development to clinically deployable surgical autonomy has not been comprehensively synthesised. This PRISMA 2020-guided systematic review examines RL, IL, safe RL, simulation-to-real (sim-to-real) transfer, foundation models, VLA systems, and regulatory readiness in surgical robotics. We searched IEEE Xplore, PubMed/MEDLINE, Embase, Scopus, Web of Science, the Cochrane Library, ACM Digital Library, arXiv, and medRxiv for studies published between January 2015 and March 2026, with additional studies identified through backward citation tracing. Eligible studies proposed novel RL, imitation learning, or foundation-model approaches for surgical robotics with empirical validation in simulation or on physical robotic platforms. Two reviewers independently extracted data using a predefined coding scheme, and a third reviewer resolved disagreements. Owing to substantial heterogeneity in platforms, tasks, and outcome measures, a quantitative meta-analysis was not feasible; therefore, the evidence was synthesised narratively using a comparative framework. A total of 220 studies met the inclusion criteria, covering eleven active surgical RL platforms, seven paired sim-to-real studies, emerging foundation-model architectures, and three FDA-cleared robotic systems exhibiting Level 3 autonomy. Available comparative studies suggest that hierarchical approaches can outperform flat policies in long-horizon tasks, while language-conditioned models demonstrated promising multi-step surgical capabilities. Seven paired simulation-to-real studies were identified, encompassing tissue retraction, guidewire navigation, and surgical cutting tasks. Sim-to-real performance gaps varied substantially by task and metric, with success-rate gaps ranging from −10 to 50 percentage points (negative values indicating better real-world than simulated performance), while paired mean spatial errors differed by at most 0.61 mm. Most studies employed domain randomization or visual domain adaptation; hierarchical reinforcement learning demonstrated advantages over flat policies in multi-step surgical tasks. Explicit safety-constrained methods (CPO, CBF, and SER), formal verification, and regulatory-aligned evaluation were reported in fewer than 3% of applied studies. Most evidence remained simulation-based, with no reported autonomous RL execution in vivo in humans. Overall, RL-based surgical robotics appears mature at the simulation stage but remains preclinical for autonomous clinical deployment. Future progress requires stronger sim-to-real validation, multimodal safety-aware architectures, alignment with IEC 62304, ISO 14971, FDA guidance, and the EU AI Act, and open benchmarks that jointly evaluate performance, safety, and surgeon trust. Full article

Robust three-finger grasping under physical-domain variation remains challenging because contact stability can change substantially with object mass, effective friction, and observation noise. This work develops U-GRA, a conservative offline-to-online residual adaptation framework for simulated three-finger grasping. U-GRA introduces a unified prior-preserving and critic-disagreement-regulated architecture that couples a frozen behavioral prior with a spectrally normalized and bounded residual stream, scalar Twin-Q reliability assessment, and critic-conditioned residual fusion. The framework first learns a nominal behavioral prior from successful demonstrations and then freezes it as a stable action anchor during online adaptation. Before execution, the twin critics evaluate a candidate action formed from the prior action and the bounded residual proposal, and their absolute scalar Q-value disagreement conditions a state-dependent gate that regulates residual-injection strength. Experiments are conducted in CoppeliaSim using an offline dataset of 40,000 successful demonstrations and online randomization of object mass, effective friction, and observation noise. Across three independent seeds, U-GRA achieves a mean success rate of $84.8\pm 2.3\%$, a normalized return of $82.7\pm 4.1$, and a jitter value of $0.12\pm 0.03$. Relative to AWAC-Res, the strongest evaluated baseline, U-GRA improves mean success by 9.2 percentage points and reduces jitter by 57.1%. It also retains the highest mean success rate and normalized return over the unseen simulated high-mass–low-friction OOD region. These results provide simulation evidence that preserving a nominal behavioral prior while regulating bounded residual correction through critic disagreement improves three-finger grasping robustness under physical-domain variation. Full article

Climate change education has expanded in curricular, institutional and media arenas, yet it still faces a central challenge: turning scientific knowledge into educational practices that transform perceptions, decisions and participation. This article argues that the challenge cannot be addressed by merely adding content about the climate system, because the climate crisis is experienced in specific territories, affects communities unequally and activates emotions that can either enable or inhibit action. Through a critical integrative review and research-reflection approach, the article connects three bodies of scholarship that are commonly developed in parallel: territorial and place-based learning, climate justice, and emotional literacy. Its conceptual novelty lies not in claiming that these components are individually new, but in specifying territorial memory as the mediating mechanism through which scientific evidence, unequal vulnerability, affective experience and collective action are combined in teacher education. On this basis, it presents the TEJA Model—territorialization, experience, justice and action—as a heuristic framework for designing, analyzing and evaluating teacher-education sequences. Its practical contribution is a five-phase process and an operational matrix that translate nearby risks such as floods, droughts and wildfires into inquiry, deliberation, emotional care and feasible collective action. The model is presented as a testable conceptual proposal rather than a validated intervention. The article concludes that relevant climate education must be scientific, human, territorial and political: it should recognize vulnerability, begin from lived places and address unequal responsibilities and collective capacities for action. Full article

The rapid integration of Generative Artificial Intelligence (GenAI) technologies into educational environments has generated new opportunities for developing intelligent systems capable of monitoring learner interactions, modeling learning behaviors, and supporting adaptive educational decision-making. As learners increasingly engage with AI-powered tools for content generation, information retrieval, problem solving, and knowledge construction, educational platforms require robust analytics architectures that can transform human–AI interaction data into actionable insights for instructors, administrators, and learning support systems. This paper proposes a computational architecture for learning behavior analytics in AI-enhanced educational environments. The architecture integrates multiple analytical components, including learner interaction monitoring, behavioral data aggregation, AI utilization profiling, performance-related indicator analysis, and decision-support modules for adaptive intervention and learner support. The proposed framework is designed to capture measurable dimensions of AI-assisted learning behavior, enabling educational systems to identify usage patterns, model learner engagement, and generate analytics-driven recommendations for instructional improvement. The architecture adopts a data-driven approach in which behavioral indicators derived from learner interactions with GenAI tools are processed through learning analytics mechanisms to support predictive modeling, learner classification, and intelligent feedback generation. The framework further incorporates dashboards and reporting components that facilitate real-time monitoring of AI-assisted learning activities and provide evidence-based insights for educational stakeholders. To demonstrate the applicability of the proposed architecture, a pilot implementation was conducted using learner interaction and perception data collected from higher education students. Preliminary analytical results indicate that task-specific AI utilization behaviors provide meaningful behavioral signals that can be incorporated into learner modeling and adaptive learning analytics processes. These findings support the feasibility of integrating GenAI interaction data into intelligent educational systems for monitoring and decision-support purposes. The proposed architecture contributes to the development of next-generation educational technologies by providing a scalable framework for learning behavior analytics, human–AI interaction modeling, and intelligent educational decision support. The study offers practical implications for the design of adaptive learning platforms, educational data analytics systems, and AI-enabled learning environments that support effective and responsible human–AI collaboration. Full article

Smart city and smart region governance increasingly relies on data-informed decision-making, stakeholder participation, digital tools, and public feedback. However, these mechanisms do not automatically create institutional learning. This article examines sustainable smart region governance as an institutional learnability problem and asks whether participation, local knowledge, and feedback are converted into adaptive action and retained as institutional memory. The study applies a Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR)-informed structured evidence mapping combined with an embedded regional case diagnosis from the Tokaj Wine Region, Hungary. The analysis integrates the literature on smart governance, learning regions, higher education quality assurance, territorial resilience, and adaptive governance with regional governance observation materials, stakeholder survey data, coding tables, calculation workbooks, and analytical figures deposited in a public Figshare dataset. The results identify the Act/Remember gap as the central learning-cycle disruption. Planning, implementation, monitoring, and consultation may be present, but feedback often fails to become adaptive action, and action is weakly retained as institutional memory. The comparison with higher education quality assurance shows that structured feedback and continuous improvement principles are transferable only as learning logic, not as procedural models. The findings also show that single-profile territories are especially vulnerable to delayed learning, strategic lock-in, and weak community knowledge-building. The article contributes to smart governance research by proposing institutional learnability as a diagnostic capacity of sustainable smart regions. It argues that digital tools should function as learning infrastructure supporting traceability, feedback-to-action mechanisms, and institutional memory, rather than as substitutes for human deliberation, trust, and collective responsibility. Full article

Diffusion policies model multimodal robot action sequences, but behavioral cloning does not directly optimize task return. We present a structured scoping review of reinforcement learning for generative robot policies and a bounded state-based locomotion reproduction. Four documented routes yielded 178 records, 162 unique candidates, and an 84-study evidence map. Hierarchical rules distinguish 41 direct reward-driven studies from 32 adjacent robotic, eight alternative-generator, and three non-robotic studies; a five-axis taxonomy codes initialization/data, interaction regime, optimized object, credit assignment, and generator. Under a fixed-final evaluation protocol on the Datasets for Deep Data-Driven Reinforcement Learning (D4RL) 1.1 Hopper benchmark, five diffusion policy policy optimization (DPPO) fine-tuning seeds improved over their run-recorded behavior-cloning initializations by a mean of 1261.2 return, with a seed-level standard deviation of 125.5 and a 95% confidence interval of 1105.3–1417.1; the five runs link to two recorded behavior-cloning checkpoints. A Gaussian-policy control also improved after proximal policy optimization, so the gain was not diffusion-specific. A full-chain backpropagation adaptation exhibited clear seed-dependent variation, a matched action-divergence intervention did not establish causal critical timesteps, and reducing denoiser evaluations from 20 to 2 lowered A100 latency from 30.97 to 3.85 ms while substantially reducing normalized score. The experiments are limited to state-based locomotion and do not validate visual manipulation. Full article

Digital transformation has fundamentally reshaped higher education, creating a need for adaptive systems capable of integrating technological infrastructure, learner capabilities, institutional readiness, and learning analytics into a unified decision-support environment. While existing studies predominantly examine online learning from pedagogical or behavioural perspectives, comparatively little attention has been devoted to configurable system architectures that support institutional monitoring and continuous optimization of digital learning ecosystems. This paper addresses this gap by proposing a configurable Digital Learning Capability Assessment Framework (DLCAF), a modular systems framework designed to assess, monitor, and optimize digital learning environments through the integration of infrastructure capabilities, digital literacy, learner motivation, technology acceptance, and institutional performance indicators. The framework employs a layered architecture comprising data acquisition, capability assessment, analytics, decision-support, and feedback modules, enabling flexible configuration according to institutional requirements and educational contexts. To demonstrate the applicability of the proposed framework, a survey involving 220 students from 27 study programs across Croatian higher education institutions was conducted during the COVID-19 digital transition. The empirical findings serve as an application case for validating the framework and illustrating how learner perceptions, technical constraints, institutional support, and digital readiness can be systematically incorporated into an adaptive decision-support process. The results indicate that technical infrastructure, learner motivation, digital competencies, communication quality, and institutional support collectively influence the effectiveness of digital learning environments. The proposed framework transforms these heterogeneous indicators into actionable institutional intelligence that supports evidence-based planning, continuous monitoring, and targeted intervention strategies. By repositioning digital learning evaluation as a systems engineering problem rather than solely an educational assessment exercise, this work contributes a reusable and extensible framework that can be deployed across diverse higher education environments. The architecture provides a foundation for future integration of artificial intelligence, learning analytics, predictive modelling, and adaptive recommendation mechanisms, supporting the development of intelligent digital learning ecosystems capable of continuous improvement and institutional decision support. Full article