The decarbonization of energy systems requires efficient and reliable high-temperature thermochemical energy storage (HT-TCES) materials to stabilize renewable electricity supply and meet industrial heat needs. Finding the best HT-TCES materials is a complex multi-criteria decision-making (MCDM) problem, as it involves balancing thermodynamic performance, thermal stability, cost, and environmental sustainability among many potential options. This study presents a machine learning-integrated multi-criteria decision-making framework for the methodical selection of HT-TCES materials. Eighteen material alternatives are rigorously evaluated based on eight crucial criteria, including reaction enthalpy, energy density, operating temperature range, cycle stability, heat transfer characteristics, raw material cost, environmental impact, and scalability. The proposed method uses Multiple-Criteria Ranking by Alternative Trace (MCRAT) as the main ranking tool, supported by three weighting strategies—Analytical Hierarchy Process (AHP) for expert-driven prioritization, CRITIC (Criteria Importance Through Intercriteria Correlation) for objective variability assessment, and MEREC (Method based on the Removal Effects of Criteria) to ensure robustness. A random forest machine learning algorithm verifies the MCDM rankings, identifies non-linear relationships among criteria, and conducts sensitivity analysis to identify the most impactful parameters. The integrated ML-MCDM approach provides consistent, data-based material rankings and highlights key properties that determine HT-TCES suitability. This hybrid framework offers a reproducible and scalable decision-support tool to accelerate the deployment of advanced thermochemical storage systems, fostering improved grid flexibility, industrial decarbonization, and wider adoption of renewable energy.

The pharmaceutical and nutraceutical industries face the challenge of optimizing resources and reducing waste. In this context, the valorization of by-products from agro-industrial fruit and vegetable processing represents a sustainable and economically relevant strategy. Inadequate disposal of these co-products generates significant environmental and economic impacts, threatening food security.

This study aimed to evaluate the nutritional and bioactive potential of pumpkin peel to propose a valorization strategy as a food additive. We conducted a comprehensive characterization of its chemical composition, including the quantification of total sugars, lipids, vitamins, mineral salts, and ash. Simultaneously, the content of bioactive compounds (phenolic compounds, carotenoids, and flavonoids) was determined. Antioxidant activity was assessed by a battery of in vitro tests, including total antioxidant capacity (TAC), DPPH and ABTS radical scavenging tests, ferric reducing antioxidant power (FRAP), and the β-carotene bleaching test. Furthermore, antidiabetic activity was explored.

The obtained results demonstrate that pumpkin peel is a substantial source of vitamins, mineral salts, and bioactive phytochemicals, notably carotenoids and phenolic compounds. This richness confers significant antioxidant and antidiabetic properties to this by-product. These findings suggest a high potential for integrating pumpkin peel into food formulations, particularly those intended for nutritional enrichment and the prevention or management of diabetes, thereby contributing to a circular and sustainable approach in the agro-industry.

Brain tumors are abnormal cell masses in the brain. For early-stage tumor classification based on neuroimaging techniques, particularly MRI, DL approaches such as DNNs and CNNs are frequently employed and have achieved moderate success. Recently, to address limitations of classical AI—such as data-driven challenges (e.g., high complexity, correlations) and limited computational resources (CPU, GPU)—quantum computing-based AI approaches have been developed, leveraging quantum mechanics principles and properties of quantum particles. This study proposes a hybrid quantum–classical integrated neural network (HQCINN) model for multi-class brain tumor classification, comparing its performance with two DNN and CNN models, with trainable parameters controlled at a similar level. The HQCINN model consists of quantum and classical components: the quantum part incorporates amplitude encoding, a multi-layer parameterized quantum circuit, and measurement operations, while the classical part includes the softmax function, loss computation, and optimization steps. Furthermore, the proposed quantum model was executed on the default.qubit state-vector simulator provided by PennyLane 0.35.1, whereas Keras was used for designing the DL models. The HQCINN model demonstrated the highest performance in distinguishing four different brain tumor types, with training/validation losses of 0.24/0.23 and accuracies of 0.91/0.92. For the DNN model, the losses were 1.31/1.03 and accuracies 0.47/0.51, while for the CNN model, losses were 0.42/0.80 and accuracies 0.78/0.65. The total execution time for HQCINN was 16 hours longer than that of the CNN. In conclusion, model selection should be scenario-dependent: HQCINN offers superior performance on complex datasets, whereas classical CNNs remain more efficient when speed is prioritized.

Numerous studies have highlighted the strong association between reduced incidence of chronic diseases and diets rich in fruits and vegetables, which are abundant sources of bioactive compounds. Among these, carotenoids have garnered significant attention for their potential to modulate and prevent metabolic dysfunctions. In this study, we investigated the nutraceutical potential of a β-cryptoxanthin-rich carotenoid extract obtained from persimmon (Diospyros kaki Thunb.) by-products. High-performance liquid chromatography (HPLC) confirmed the carotenoid profile of the extract, with β-cryptoxanthin as the major compound. In vitro assays using 3T3-L1 murine adipocytes demonstrated that the extract reduced reactive oxygen species (ROS) and intracellular triglyceride accumulation without cytotoxic effects. In vivo, C57BL/6J mice were fed either a high-fat diet (HFD) or a low-fat diet (LFD), both with or without extract supplementation, over 12 weeks. The extract significantly reduced serum triglycerides, total cholesterol, LDL-cholesterol, and pro-inflammatory cytokines, while improving HDL-cholesterol levels. Hepatic tissue analysis revealed a marked reduction in collagen deposition and lipid accumulation in HFD-treated mice, indicating hepatoprotective effects. Furthermore, in silico docking suggested a potential interaction of β-cryptoxanthin with ADIPOR and GLUT1 receptors, possibly underlying the observed metabolic improvements. Collectively, these findings support the application of β-cryptoxanthin-rich persimmon extract as a promising nutraceutical strategy to combat obesity and metabolic-associated fatty liver disease (MAFLD). Further studies are warranted to confirm its efficacy in clinical settings.

This work proposes a home-centric monitoring system that integrates machine vision, YOLO-based object detection, and household sensing platforms, fixed indoor cameras and mobile robots such as robot vacuums, to track the activities and partial affective indicators of indoor cats. The system targets scenarios in which owners are away for work or travel and is especially relevant for multi-cat households, where social dynamics, resource access, and routine changes can influence wellbeing. By operating primarily on-device, the approach aims to deliver real-time insights while preserving privacy and minimizing bandwidth.

The architecture combines YOLO for fast detection of cats and relevant objects, multi-object tracking and re-identification for per-cat continuity, and posture and action recognition for behavior classification. Temporal analytics convert frame-level outputs into interpretable daily and weekly metrics, including locomotion patterns, play bouts, resource visits, zone occupancy, and trend deviations from individualized baselines. The mobile robot augments fixed viewpoints by patrolling occluded areas and following up on uncertain detections, improving coverage and robustness in cluttered indoor environments.

To estimate partial emotion-related indicators, the system aggregates proxies such as posture cues (ear position, tail carriage, body curvature), activity balance, hiding or exploration patterns, and resource interaction changes, optionally fused with audio or smart-home signals. It highlights potential stress or discomfort when multi-cue deviations persist, while emphasizing that such inferences are approximations and not medical diagnoses. Deployment guidance addresses model selection for edge devices, dataset curation tailored to the home, per-cat baselining, and owner-in-the-loop corrections.

Autonomous lunar probe landing presents a complex control challenge due to limited sensor feedback, delayed communications, and dynamic terrain conditions. In this work, we present a lightweight and optimized reinforcement learning solution using a Deep Q-Network (DQN) agent trained on the LunarLander-v3 environment from the Gymnasium library. Our aim is to develop a model capable of precise, resource-efficient landings under constrained simulation settings.

The agent interacts with an 8-dimensional state space and 4 discrete action choices, learning through experience replay and an ε-greedy policy. We systematically evaluated the impact of neural network architecture (Tiny, Base, Wide, Deep) and conducted extensive hyperparameter tuning via grid search across learning rates, discount factors, and update rates. The best-performing configuration—128-128 Wide architecture, learning rate 0.0005, discount factor 0.99, soft update rate 0.01—demonstrated superior performance, achieving an average reward of 262.89 in 355.98 seconds of training.

Final testing revealed that training to a threshold of 250 reward yields a 93% landing success rate, outperforming both under-trained and over-trained agents in terms of efficiency and generalization. This result was validated across 100 test episodes, confirming consistent, high-accuracy autonomous landing behavior.

Our findings highlight the viability of deploying lightweight, well-tuned DQN agents for real-time lunar landing scenarios. The proposed approach serves as a scalable blueprint for future space robotics systems, bridging the gap between simulation and real-world feasibility. Future work will incorporate terrain complexity and uncertainty modeling to extend robustness in dynamic planetary environments.

Autonomous navigation in hazardous environments demands path planning strategies that balance computational efficiency with safety considerations. This study compares the performance of three widely used algorithms—A*, D*, and Rapidly Exploring Random Trees (RRTs)—across varying risk conditions. A grid-based framework was employed to simulate three types of environments: mixed-risk scenarios with randomly distributed obstacles, a fully high-risk environment, and a fully low-risk environment. Performance was assessed using execution time, path length, and collision behavior as evaluation metrics. Results demonstrate that A* consistently achieves the fastest execution across all scenarios, confirming its computational efficiency. However, in fully high-risk and low-risk environments, A* tends to generate longer paths compared to RRT. While RRT frequently identifies shorter and more economical paths, its sampling-based approach results in longer computation times than A* and D*, and in some cases, introduces instability in path safety. D* shows performance similar to A* in terms of path length but with slightly higher computation time. Overall, A* emerges as the most reliable option for time-critical applications, whereas RRT offers path-length advantages at the expense of speed and stability. The findings highlight the trade-offs between graph-based and sampling-based methods and suggest that hybrid or risk-aware planners may provide more robust solutions for real-world rescue, surveillance, and hazardous material handling scenarios.

Due to the diverse cultural backgrounds and varying cognitive levels of foreign English language learners, personalized scaffolding is essential for developing language competency across all four language skills: reading, writing, listening, and speaking. However, this need is not adequately addressed in the existing literature. In the post-method era, glocal (global and local) teaching and learning are encouraged, where learners’ cultural foundations and cognitive stages help define the learning approach. In this context, while human scaffolding is limited, the advancement of Generative AI, in English as a Foreign Language (EFL) education, promotes personalized learning and fosters learner autonomy in both formal and informal settings. Grounded in Constructivist theory, which emphasizes contextualized learning, GenAI is integrated as a scaffold to enhance the Zone of Proximal Development (ZPD). Guided by the PICO model and PRISMA screening process, this study selected 45 secondary documents for coding from an initial pool of 383, sourced from SCOPUS through a Boolean search. A thematic qualitative approach was employed for analysis. Our findings reveal that a single method is not sufficient and that GenAI tools play a crucial role in navigating contextual learning pathways, enhancing students' engagement and learning outcomes in English language learning. These tools improve proficiency across all four language skills and reflect a shift from method-bound instruction to personalized, technology-enhanced learning. The findings will assist students, teachers, and curriculum designers in redefining their understanding of English language acquisition by examining how AI can transform English education from a traditional, teacher-controlled model to a learner-driven system.

This work presents an alternating sequential model predictive control (ASMPC) scheme applied to multimodular direct matrix converters (MMDMCs). A common limitation of model predictive control (MPC) in power converters is the need to assign weighting factors when several objectives are considered simultaneously. In practice, choosing appropriate weights often requires empirical adjustment and depends strongly on the system and operating conditions. Even sequential MPC (SMPC), which addresses part of this problem by prioritizing objectives, still implicitly relies on weighting through its predefined hierarchy and may lead to unbalanced performance when conditions change.

To address these deficiencies, the proposed ASMPC alternately evaluates two control objectives—load current tracking and input reactive power minimization—at each sampling step, without requiring explicit weighting factors or fixed priority ordering. This strategy was implemented in MATLAB/Simulink using an architecture composed of two direct matrix converters operating in parallel. The influence of parameter N₂, which defines the number of candidate states considered after the first evaluation, was analyzed under step changes in reference currents of 30 A and 60 A. Performance metrics such as total harmonic distortion (THD) and mean squared error (MSE) were evaluated, supported by a descriptive statistical analysis including mean, standard deviation, mean absolute deviation (MAD), and coefficient of variation (CV).

Simulation results show low dispersion and stable performance against variations in N₂. Overall, ASMPC provides a robust, efficient, and easy-to-implement solution for modular power converter systems, showing clear advantages over classical and sequential predictive control schemes.

This review provides a comprehensive overview of recent advances in the optimization and performance analysis of lead-free and inorganic perovskite solar cells, with a particular focus on CsSnCl₃, CsPbI₃, and Cs₂CuBiBr₆. Using insights derived from SCAPS-1D-based studies, this review highlights comparative trends between lead-based and lead-free perovskites, emphasizing the environmental advantages and emerging efficiency potential of non-toxic alternatives. Reported simulation data indicate that optimized device architectures—such as ITO/CeO₂/CsSnCl₃/CBTS/Au and ITO/TiO₂/CsPbI₃/CBTS/Au—exhibit distinct variations in photovoltaic parameters. Specifically, CsSnCl₃ achieves a Voc of 0.96 V, Jsc of 32.22 mA/cm², FF of 86.40%, and a PCE of 25.1%, whereas CsPbI₃, a lead-based perovskite, delivers a Voc of 0.997 V, Jsc of 21.07 mA/cm², FF of 85.21%, and a PCE of 17.9%. For Cs₂CuBiBr₆, a lead-free double perovskite, efficiencies range from 18.69% to 19.70%, depending on the selected electron transport layer. These results collectively demonstrate that lead-free materials can achieve competitive or even superior performance while offering improved environmental compatibility. Furthermore, the review discusses the influence of absorber thickness, ETL/HTL combinations, series and shunt resistances, and temperature on overall device operation. Overall, this analysis underscores the growing potential of lead-free perovskites as sustainable, high-efficiency alternatives fornext-generation thin-film photovoltaic technologies.