Search Results (122)

Shared GPU clusters often execute multiple distributed training jobs concurrently under fluctuating contention. We reinterpret this setting as a two-scale control problem, where the micro scale captures intra-job learning dynamics and the macro scale captures inter-job resource arbitration. We propose an entropy-guided hierarchical framework that links these two scales through a unified uncertainty signal computed from training logits. Unlike existing uncertainty-aware methods that typically use uncertainty for only a single level of decision making, our approach uses the same entropy-based signal to jointly support both intra-job adaptation and inter-job scheduling within a hierarchical control loop. At the micro level, each worker estimates predictive uncertainty via normalized entropy and converts it into stable weights that drive epoch-level controls for uncertainty-aware data sharding, fixed-budget batch-size reallocation, and learning-rate modulation, while remaining compatible with standard synchronous data-parallel training. At the macro level, the same signal is aggregated into a job utility score that guides admission, ordering, and GPU quota assignment under contention. In large-scale workload-driven simulation, our method reduces average job completion time (JCT) by 23.7% and shortens cluster makespan by 15.7% relative to a strong learning-unaware baseline, demonstrating that uncertainty-aligned scheduling can improve cluster-level efficiency while preserving training correctness. We further validate scalability using a calibrated simulator up to 1024 nodes. Full article

Saudi Arabia experienced rapid convergence in women’s financial inclusion between 2014 and 2024, a period marked by the 2018–2019 reforms expanding women’s economic rights and the accelerated deployment of digital payment infrastructure. Using four waves of Global Findex microdata (2014, 2017, 2021, and 2024), this study estimates probability-weighted logit models with average marginal effects and decomposes gender gaps using nonlinear Kitagawa and Blinder–Oaxaca methods. Reform-era dynamics are examined by tracing changes in the gender gap across survey waves. The findings indicate that aggregate gender gaps in account ownership and digital payment usage narrowed substantially by 2024, with conditional gaps among employed adults no longer statistically significant, while sizable disparities persist among individuals outside the workforce. Decomposition results highlight increased female labor force participation as a key correlate of convergence, consistent with labor market integration playing a central role in women’s financial inclusion during the reform era. Full article

Real-time route guidance during disruptions in urban rail transit systems requires rapidly providing effective strategies that simultaneously alleviate congestion and account for passengers’ travel time. This study proposes an optimization framework that considers travel time, congestion perception time, and information costs, incorporating a Logit choice model with information bias to reflect passengers’ behavioral responses under disruptions. A bi-level simulation evaluation mechanism is employed to rapidly evaluate the objective functions under different guidance strategies, where a Physically Consistent Incremental Simulator, based on differential computation, achieves a 599-fold speedup while maintaining high fidelity with full-scale simulations (Pearson correlation > 0.96). A hybrid algorithm combining the Gray Wolf Optimizer and Adaptive Large Neighborhood Search is developed to solve the origin–destination level route guidance optimization problem. The algorithm embeds domain knowledge-based “destroy and repair” operators with a sequential repair mechanism to enable fast global search and precise local refinement. Case study results demonstrate that the framework reduces severely congested sections by 36%, shortens average travel time by 7.16 min, and improves solution quality by 12–30% over baseline algorithms. These findings confirm the practical applicability of integrating intelligent optimization with high-efficiency simulation for emergency route guidance in large-scale metro networks. Full article

Background: The Glycemic Index (GI) serves as a critical indicator of carbohydrate quality linked to postprandial glycemic response. As “Low-GI” claims proliferate on front-of-pack labels, it remains unclear how consumers value this complex signal. This study quantifies willingness to pay (WTP) for Low-GI labeling and tests a “motivation–capability” mechanism, positing that health orientation motivates label use, while objective Low-GI knowledge facilitates targeted evaluation across nutritional contexts. Methods: A discrete choice experiment was conducted in China using plain yogurt (N = 910). Mixed logit models analyzed how the valuation of the Low-GI claim is moderated by carbohydrate context, health orientation, and objective knowledge. Results: Results indicate a significant average premium for Low-GI labeling, with health orientation acting as a consistent motivational amplifier. Objective knowledge functions as a critical moderator interacting with carbohydrate context, driving label valuation only in specific low- or high-carbohydrate profiles while triggering skepticism in regular carbohydrate ones. Conclusions: These findings suggest that the public health effectiveness of emerging physiological claims depends jointly on consumer motivation and label-specific literacy. Consequently, policy interventions should combine label standardization with targeted education, equipping consumers with the capability to decode the claim’s physiological meaning rather than relying on a generalized health halo. Full article

High-quality, interoperable biodiversity information is a prerequisite for effective conservation policy, compliance with European Union (EU) reporting obligations, and efficient environmental decision-making. Greece’s LIFE EL-BIOS (LIFE20 GIE/GR/001317) developed the first National Biodiversity Information System, aiming to aggregate, standardise, and disseminate spatial and non-spatial data for species, habitats, pressures, and trends. This paper provides an economic valuation of this information system as a public, non-market good. We designed a two-stage stated-preference study: (i) a short pre-survey to calibrate initial bids and (ii) the main survey employing double-bounded dichotomous choice (DBDC) contingent valuation with a spike-logit specification. The payment vehicle was a hypothetical monthly subscription in a post-LIFE scenario. The instrument measured time savings (hours), perceived reliability (Likert 1–5), and key demographics/roles. A total of 167 valid responses were collected in September 2025. Participants reported an average of 5.2 h saved per use (median 4; max 14). Among those expressing willingness to pay (WTP), 77% rated EL-BIOS reliability as “High/Very high”. Econometric results indicate time savings as the strongest positive determinant of WTP; perceived reliability is positive and marginally significant; years of experience are negatively associated with acceptance; and cost has a strong negative effect. Mean WTP is estimated at €6.7 per month (median €3.5). Notably, 64% of those unwilling to pay declared protest motives (data should remain public and free). Accordingly, non-payment is decomposed into true zero WTP versus protest-based refusal, i.e., refusal to pay despite acknowledging value. This high protest share reflects principled opposition to paying for public biodiversity data rather than low perceived value of the system. The EL-BIOS database generates measurable productivity gains and social value both through positive WTP and principled protest responses supporting open public data. These findings inform policy on sustainable financing, governance, and long-term operation of national biodiversity information systems. Full article

This study presents a comprehensive comparative analysis of the effect of route set size on stochastic user equilibrium (SUE) traffic assignment, focusing on both logit-based (Multinomial Logit (MNL) and Path Size Logit (PSL)) and weibit-based models (Multinomial Weibit (MNW) and Path Size Weibit (PSW)). The primary objective is to investigate the influence of route set size on traffic patterns and determine the minimum requisite number of routes for flow stabilization within the SUE framework. The analysis, conducted on the Winnipeg network using a customized Self-Regulated Averaging (SRA) scheme, yields three key findings. First, all models successfully converged, but the weibit-based models (MNW and PSW) converged faster than the logit-based models. Second, an analysis of perceived total travel time demonstrated that the majority of efficiency gains from route inclusion diminish after a threshold of approximately maximum 30 routes to 40 routes per O-D pair, indicating this number is sufficient for achieving stable SUE results in both model families. Third, the weibit-based model was found to be more sensitive to route overlap effects, continuing to adjust flow patterns up to maximum 45 routes per O-D pair, and exhibiting a greater tendency to allocate flow to less overlapping outer roads. This highlights the superior capability of the weibit formulation, which accounts for heterogeneous perception variance, to achieve a more behaviorally realistic equilibrium compared to the logit models. Full article

Using a unique multi-wave dataset from nationally representative surveys in Portugal (2015, 2020, and 2023), this study extends the household finance literature by examining the mechanisms linking financial literacy to capital market participation. We propose and test a moderated mediation framework, arguing that the relationship is channeled through the mediating roles of financial resilience and self-efficacy and is contingent upon sociodemographic moderators. Our findings reveal a decline in average financial knowledge between 2015 and 2020/23, with persistent gaps across socioeconomic groups. Empirical results from count, logit, and ordered logit models provide strong evidence for partial mediation; financial literacy significantly enhances a household’s financial resilience, which in turn is a strong positive predictor of participation in stocks, bonds, and mutual funds. Furthermore, we find that perceived financial knowledge is a more powerful direct driver of participation than objective knowledge. Crucially, these pathways are powerfully moderated by income and education, highlighting that socioeconomic status is a fundamental boundary condition for converting knowledge into investment behavior. The results challenge simplistic direct-effects models and suggest that policy initiatives aimed at boosting market participation, such as the Portuguese National Plan for Financial Education, must look beyond knowledge dissemination to also foster financial resilience, self-efficacy, and address structural inequalities. Full article

Geothermal energy provides a stable baseload renewable source that is less affected by weather variability compared with solar and wind power, and is therefore increasingly considered in national energy transition and net-zero strategies. Yet its environmental externalities and associated social benefits are not fully priced in existing electricity markets, raising the question of how much the public is willing to pay for geothermal-based generation. This study applies non-market valuation theory to estimate citizens’ additional annual electricity payment required to replace coal-fired generation with geothermal energy. A contingent valuation method (CVM) survey was conducted through face-to-face interviews, employing a closed-ended single-bounded dichotomous choice format with incentive compatibility. Stratified random sampling yielded 678 valid observations. The estimated mean willingness to pay (WTP) per person per year is USD 56.18 (NTD 1792) under the Probit model and USD 52.16 (NTD 1663) under the Logit model, representing approximately 0.2–0.3% of average annual income and 16–20% of the average annual electricity bill. Aggregated to the population level, total annual WTP amounts to USD 688 million (NTD 21,934 billion; Probit) and USD 638 million (NTD 20,355 billion; Logit). These estimates correspond to support for developing approximately 108–335 MW of geothermal capacity, sufficient to supply around 202,000–624,000 four-person households. The findings indicate substantial public support for geothermal power as part of Taiwan’s renewable energy transition, and provide empirical evidence relevant to regions with comparable geothermal potential. Full article

This paper presents a novel dual-stream framework for multimodal emotion recognition, engineered to address the varying complexity inherent in emotional expressions. The proposed architecture uniquely integrates a graph embedding as an auxiliary modality to explicitly model temporal correlations between utterances, and processes features through two complementary sub-models operating in parallel: a Cross-Attention Transformer Mixture-of-Experts (MoE) model and a Sum-Product Linear MoE model. The former deciphers nuanced and ambiguous emotions, such as ‘fear’ and ‘sadness’, by leveraging a deep cross-attention mechanism to model intricate, bidirectional dependencies between textual and acoustic features. The latter is a lightweight model optimized to efficiently recognize clear and intuitive emotions, like ‘anger’ and ‘happiness’, through simple element-wise operations. The final prediction is derived from an average ensemble of logits from both sub-models, ensuring a robust and balanced classification. Evaluated on the Korean AI-Hub and English IEMOCAP datasets, the framework achieves state-of-the-art accuracy of 0.8071 and demonstrates excellent cross-lingual generalization with an accuracy of 0.7823. Empirical results validate the complementary design, confirming that the specialized models synergistically enhance overall performance. Full article

Object detection is an important tool in many areas, such as robotics or autonomous driving. Especially in these areas, a wide variety of object classes must be detected or interacted with. Models from the field of Open-Vocabulary Detection (OVD) provide a solution here, as they can detect not only base classes but also novel object classes, i.e., those classes that were not seen during training. However, one problem with OVD models is their poor calibration, meaning that the predictions are often too over- or under-confident. To improve the calibration, Temperature Scaling is used in this study. Using YOLO World, one of the best-performing OVD models, the aim is to determine the optimal T-value for this calibration method. For this reason, it is investigated whether there is a correlation between the logit distribution and the optimal T-value and how this can be modeled. Finally, the influence of Temperature Scaling on the Expected Calibration Error (ECE) and the mAP (Mean Average Precision) will be analyzed. The results of this study show that similar logit distributions of different datasets result in the same optimal T-values. This correlation could be best modeled using Kernel Ridge Regression (KRR) and Support Vector Machine (SVM). In all cases, the ECE could be improved by Temperature Scaling without significantly reducing the mAP. Full article

Background: Human immunodeficiency virus (HIV) affects more than 39 million people worldwide, with Uganda ranked 10th among countries with the highest number of cases. As new preventative HIV injectables emerge, it is vital to think about how best to tailor strategies to promote these injectable drugs, like PrEP and vaccines, when available, to the different populations most in need. Discrete choice experiments (DCEs) are economics-derived methods used to determine factors that influence engagement in a certain behavior. Objective: This study used a DCE to determine the preferences for a preventative HIV injectable drugs/vaccines among people at risk of HIV acquisition in urban and peri-urban areas of Uganda. Methods: In June 2024, we implemented a cross-sectional DCE survey in three urban sites in Uganda in English and Luganda. The survey collected information on demographics, HIV risk, vaccine confidence and responses to the 13 injection product choice tasks presented to determine preferences. We used community-based, respondent-driven sampling methods to recruit participants from three key populations: (1) female sex workers; (2) people who identify as lesbian, gay, bisexual or transgender; and (3) young women (18–24 years). We collected the data on tablets using the Sawtooth Lighthouse Studio software (v. 19.15.6), taking into consideration privacy and confidentiality, given the sensitivity of the information and recent governmental policies in Uganda. Data were analyzed using a split-sample mixed logit regression analysis. The study was approved by local ethical regulatory bodies. Results: From the total of 406 participants screened for this study, 376 participants met the eligibility criteria and were included in the final analysis (85 young women, 159 female sex workers, and 132 who identified as lesbian, gay, bisexual or transgender). The average age was 23.7 (SD: 5.7). The majority of participants had received some secondary school or vocational school (202, 53.7%) The attributes that explained the preferences were primarily severe compared to mild side effects (β: −0.69, 95% CI: −0.78, −0.60), a 30% increase in vaccine/drug effectiveness (β: 0.39, 95% CI: 0.34, 0.44), and a 50,000 UGX (or USD ~13.64) increase in cost (β: −0.22, 95% CI: −0.27, −0.17). There were no significant differences between the preferences for different injectable types. The sensitivity analyses suggested potential differences in preferences by the amount of help participants received from research assistants when completing the survey, although not by income level. Conclusions: Side effects had the greatest impact on participants’ preferences for injectable HIV prevention methods, followed closely by effectiveness and cost. It is therefore essential to develop affordable or free prevention options with minimal side effects. Policymakers should focus on reducing the financial barriers to access and emphasize transparent communication about the effectiveness and safety of these injectables in health promotion campaigns to maximize adoption and improve public health outcomes. Full article

Diagnosing dementia and recognizing substantial cognitive decline are challenging tasks. Thus, the objective of this study was to classify electroencephalograms (EEGs) recorded during a working memory task in 15 patients with mild cognitive impairment (MCogImp), 5 patients with vascular dementia (VasD), and 15 healthy controls (NC). Before creating spectrogram pictures from the EEG dataset, the data were subjected to preprocessing, which included preprocessing using conventional filters and the discrete wavelet transformation. The convolutional neural network (CNN) MobileNetV2 was employed in our investigation to identify features and assess the severity of dementia. The features were extracted from five layers of the MobileNetV2 CNN architecture—convolutional layers (‘Conv-1’), batch normalization (‘Conv-1-bn’), clipped ReLU (‘out-relu’), 2D Global Average Pooling (‘global-average-pooling2d1’), and fully connected (‘Logits’) layers. This was carried out to find the efficient features layer for dementia severity from EEGs. Feature extraction from MobileNetV2’s five layers was carried out using a decision tree (DT) and k-nearest neighbor (KNN) machine learning (ML) classifier, in conjunction with a MobileNetV2 deep learning (DL) network. The study’s findings show that the DT classifier performed best using features derived from MobileNetV2 with the 2D Global Average Pooling (global-average-pooling2d-1) layer, achieving an accuracy score of 95.9%. Second place went to the characteristics of the fully connected (Logits) layer, which achieved a score of 95.3%. The findings of this study endorse the utilization of deep processing algorithms, offering a viable approach for improving early dementia identification with high precision, hence facilitating the differentiation among NC individuals, VasD patients, and MCogImp patients. Full article

Background. Length of stay, operationalized here as nights per booking (NPB), is a first-order driver of yield, labor planning, and environmental pressure. The COVID-19 pandemic and the rise of long-stay remote workers (often labeled “slomads”, a slow-travel subset of digital nomads) plausibly altered stay-length distributions, yet national, booking-weighted evidence for the United States remains scarce. Purpose. This study quantifies COVID-19 pandemic-era and post-pandemic shifts in U.S. Airbnb stay lengths, and identifies whether higher averages reflect (i) more long stays or (ii) longer long stays. Methods. Using every U.S. Airbnb reservation created between 1 January 2019 and 31 December 2024 (collapsed to booking-count weights), the analysis combines: weighted descriptive statistics; parametric density fitting (Gamma, log-normal, Poisson–lognormal); weighted negative-binomial regression with month effects; a two-part (logit + NB) model for ≥28-night stays; and a monthly SARIMA$(0,1,1){(0,1,1)}_{12}$ with COVID-19 pandemic-phase indicators. Results. Mean NPB rose from 3.68 pre-COVID-19 to 4.36 during restrictions and then stabilized near 4.07 post-2021 (≈10% above 2019); the booking-weighted median shifted permanently from 2 to 3 nights. A two-parameter log-normal fits best by wide AIC/BIC margins, consistent with a heavy-tailed distribution. Negative-binomial estimates imply post-vaccine bookings are 6.5% shorter than restriction-era bookings, while pre-pandemic bookings are 16% shorter. In a two-part (threshold) model at 28 nights, the booking share of month-plus stays rose from 1.43% (pre) to 2.72% (restriction) and settled at 2.04% (post), whereas the conditional mean among long stays was in the mid-to-high 50 s (≈55–60 nights) and varied modestly across phases. Hence, a higher average NPB is driven primarily by a greater prevalence of month-plus bookings. A seasonal ARIMA model with pandemic-phase dummies improves fit over a dummy-free specification (likelihood-ratio = 8.39, df = 2, p = 0.015), indicating a structural level shift rather than higher-order dynamics. Contributions. The paper provides national-scale, booking-weighted evidence that U.S. short-term-rental stays became durably longer and more heavy-tailed after 2020, filling a gap in the tourism and revenue-management literature. Implications. Heavy-tailed pricing and inventory policies, and explicit regime indicators in forecasting, are recommended for practitioners; destination policy should reflect the larger month-plus segment. Full article

Ovarian cancer remains a significant global health concern, and its diagnosis heavily relies on whole-slide images (WSIs). Due to their gigapixel spatial resolution, WSIs must be split into patches and are usually modeled via multi-instance learning (MIL). Although previous studies have achieved remarkable performance comparable to that of humans, in clinical practice WSIs are distributed across multiple hospitals with strict privacy restrictions, necessitating secure, efficient, and effective federated MIL. Moreover, heterogeneous data distributions across hospitals lead to model heterogeneity, requiring a framework flexible to both data and model variations. This paper introduces HFed-MIL, a heterogeneous federated MIL framework that leverages gradient-based attention distillation to tackle these challenges. Specifically, we extend the intuition of Grad-CAM to the patch level and propose Patch-CAM, which computes gradient-based attention scores for each patch embedding, enabling structural knowledge distillation without explicit attention modules while minimizing privacy leakage. Beyond conventional logit distillation, we designed a dual-level objective that enforces both class-level and structural-level consistency, preventing the vanishing effect of naive averaging and enhancing the discriminative power and interpretability of the global model. Importantly, Patch-CAM scores provide a balanced solution between privacy, efficiency, and heterogeneity: they contain sufficient information for effective distillation (with minimal membership inference risk, MIA AUC ≈ 0.6) while significantly reducing communication cost (0.32 MB per round), making HFed-MIL practical for real-world federated pathology. Extensive experiments on multiple cancer subtypes and cross-domain datasets (Camelyon16, BreakHis) demonstrate that HFed-MIL achieves state-of-the-art performance with enhanced robustness under heterogeneity conditions. Moreover, the global attention visualizations yield sharper and clinically meaningful heatmaps, offering pathologists transparent insights into model decisions. By jointly balancing privacy, efficiency, and interpretability, HFed-MIL improves the practicality and trustworthiness of deep learning for ovarian cancer WSI analysis, thereby increasing its clinical significance. Full article

Land use change significantly affects regional carbon emissions and ecosystem service value (ESV). Under China’s Dual Carbon Goals, this study takes Beijing-Tianjin-Hebei, experiencing rapid land use change, as the study area and counties as the study unit. This study employs a combination of methods, including carbon emission coefficients, equivalent-factor methods, bivariate spatial autocorrelation, and a multinomial logit model. These were used to explore the spatial relationship between land use carbon emissions and ESV, and to identify their key driving factors. These insights are essential for promoting sustainable regional development. Results indicate the following: (1) Total land use carbon emissions increased from 2000 to 2015, then declined until 2020; emissions were high in municipal centers; carbon sinks were in northwestern ecological zones. Construction land was the primary contributor. (2) ESV declined from 2000 to 2010 but increased from 2010 to 2020, driven by forest land and water bodies. High-ESV clusters appeared in northwestern and eastern coastal zones. (3) A significant negative spatial correlation was found between carbon emissions and ESV, with dominant Low-High clustering in the north and Low-Low clustering in central and southern regions. Over time, clustering dispersed, suggesting improved spatial balance. (4) Population density and cultivated land reclamation rate were core drivers of carbon–ESV clustering patterns, while average precipitation, average temperature, NDVI, and per capita GDP showed varied effects. To promote low-carbon and ecological development, this study puts forward several policy recommendations. These include implementing differentiated land use governance and enhancing regional compensation mechanisms. In addition, optimizing demographic and industrial structures is essential to reduce emissions and improve ESV across the study area. Full article