Search Results (246)

Graph neural networks (GNNs) are deep learning models that process structured graph data. By leveraging their graphs/node classification and link prediction capabilities, they have been effectively applied in multiple domains such as community detection, location sharing services, and drug discovery. These powerful applications and the vast availability of graphs in diverse fields have facilitated the adoption of GNNs in privacy-sensitive contexts (e.g., banking systems and healthcare). Unfortunately, GNNs are vulnerable to the leakage of sensitive information through well-defined attacks. Our main focus is on membership inference attacks (MIAs) that allow the attacker to infer whether a given sample belongs to the training dataset. To prevent this, we introduce three LLM-guided defense mechanisms applied at the posterior level: posterior encoding with noise, knowledge distillation, and secure aggregation. Our proposed approaches not only successfully reduce MIA accuracy but also maintain the model’s performance on the node classification task. Our findings, validated through extensive experiments on widely used GNN architectures, offer insights into balancing privacy preservation with predictive performance. Full article

The expansion of Internet of Things (IoT) networks has enabled real-time data collection and automation across smart cities, healthcare, and agriculture, delivering greater convenience and efficiency; however, exposure to diverse threats has also increased. Machine learning-based Intrusion Detection Systems (IDSs) provide an effective means of defense, yet they require large volumes of data, and the use of raw IoT network data containing sensitive information introduces new privacy risks. This study proposes a novel privacy-preserving synthetic data generation model based on a tabular diffusion framework that incorporates Differential Privacy (DP). Among the three diffusion models (TabDDPM, TabSyn, and TabDiff), TabDiff with Utility-Preserving DP (UP-DP) achieved the best Synthetic Data Vault (SDV) Fidelity (0.98) and higher values on multiple statistical metrics, indicating improved utility. Furthermore, by employing the DisclosureProtection and attribute inference to infer and compare sensitive attributes on both real and synthetic datasets, we show that the proposed approach reduces privacy risk of the synthetic data. Additionally, a Membership Inference Attack (MIA) was also used for demonstration on models trained with both real and synthetic data. This approach decreases the risk of leaking patterns related to sensitive information, thereby enabling secure dataset sharing and analysis. Full article

Membership Inference Attacks (MIAs) pose a significant threat to privacy in modern machine learning systems, enabling adversaries to determine whether a specific data record was used during model training. Existing defense techniques often degrade model utility or rely on heuristic noise injection, which fails to provide a robust, mathematically grounded defense. In this paper, we propose Diffusion-Driven Data Preprocessing (D³P), a novel privacy-preserving framework leveraging generative diffusion models to transform sensitive training data before learning, thereby reducing the susceptibility of trained models to MIAs. Our method integrates a mathematically rigorous denoising process into a privacy-oriented diffusion pipeline, which ensures that the reconstructed data maintains essential semantic features for model utility while obfuscating fine-grained patterns that MIAs exploit. We further introduce a privacy–utility optimization strategy grounded in formal probabilistic analysis, enabling adaptive control of the diffusion noise schedule to balance attack resilience and predictive performance. Experimental evaluations across multiple datasets and architectures demonstrate that D³P significantly reduces MIA success rates by up to $42.3\%$ compared to state-of-the-art defenses, with a less than $2.5\%$ loss in accuracy. This work provides a theoretically principled and empirically validated pathway for integrating diffusion-based generative mechanisms into privacy-preserving AI pipelines, which is particularly suitable for deployment in cloud-based and blockchain-enabled machine learning environments. Full article

Traceable ring signatures (TRSs) allow a signer to create a signature that maintains anonymity while enabling traceability if needed. It merges the characteristics of traditional ring signatures with the ability to trace signers, making it ideal for applications that demand both confidentiality and accountability. In a TRS scheme, a ring of potential signers generates a signature on a message without disclosing the actual signer’s identity. However, the identity can be traced if the signer uses the same tag for multiple signatures. This paper introduces a novel formal construction of TRS under universally composable (UC) security. We integrate verifiable random functions (VRFs) and zero-knowledge proofs for membership, employing Pedersen commitments. Our signature schemes maintain a logarithmic size while preserving the UC security guarantees. Additionally, we explore the potential to extend the property of one-time anonymity in TRS to K-time anonymity. 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

Model predictive control (MPC) has been proven effective in terms of cooperative control for wind turbines (WTs). Previous work was limited to segmented linearization at a specific operating point, which significantly affected the robustness of the MPC performance. Moreover, due to nonlinearity, frequent control switching would result in the instability and fluctuation of the closed-loop control system. To address these issues, this paper proposes a novel cooperative control strategy considering fatigue load suppression for wind turbines, which is named soft switch multiple model predictive control (SSMMPC). Firstly, based on the gap metric, a model bank is constructed to divide the nonlinear WT model into several linear segments. Then, the multiple MPC is designed in a wide range of operating points. To settle the control signal oscillation problem, a soft-switching rule based on the triangular–trapezoidal hybrid membership function is proposed during controller selection. Several simulations are performed to verify the effectiveness and flexibility of SSMMPC in the partial-load region and full-load region. The results confirm that the proposed SSMMPC exhibits excellent performance in both reference operating point tracking and fatigue load mitigation, especially for the main shaft torque and tower bending load. Full article

Soil salinization, particularly under irrigation in the arid regions of North Africa, represents a major constraint to sustainable agricultural development. This study investigates the Chott El Hodna region in Algeria, a Ramsar-classified wetland severely affected by salinization. Two representative soil profiles (P1 and P2) were initially characterized, revealing chemical properties dominated by calcium-chloride and calcium-sulfate types. Based on these findings, 26 additional profiles with moderate levels of gypsum, limestone, and soluble salts were analyzed. The limited number of profiles reflects the environmental homogeneity of the area, allowing the study site to be considered a pilot zone. Fuzzy logic was employed to classify soils, identify intergrade soils, and determine their degree of membership to Solonchaks within the Calcisol class, addressing the lack of precision in conventional classifications. Results indicate that 50% of soils are Solonchaks, 46.15% are Calcisols, and 3.85% are intergrades. Principal Component Analysis (PCA) revealed that soil solution chemistry is mainly governed by the dissolution of evaporite minerals (gypsum, halite, anhydrite) and the precipitation of carbonate phases (calcite, aragonite, dolomite). Statistical analyses using Artificial Neural Networks (ANN) and Multiple Linear Regression (MLR) demonstrated that ANN achieved superior predictive performance for the Solonchak index (Is), with R² = 0.70 and RMSE = 0.17, compared with R² = 0.41 for MLR. This study proposes a robust framework combining fuzzy logic and ANN to improve the classification of saline wetland soils, particularly by identifying intergrade soils, thus providing a more precise numerical classification than conventional approaches. Full article

Stock price forecasting remains a central challenge in financial modeling due to the non-stationarity, noise, and high dimensionality of market dynamics, as well as the growing importance of unstructured textual information. In this work, we propose a multimodal prediction framework that combines FinBERT-based financial sentiment extraction with technical and statistical indicators to forecast short-term stock price movement. Contextual sentiment signals are derived from financial news headlines using FinBERT, a domain-specific transformer model fine-tuned on annotated financial text. These signals are aggregated and fused with price- and volatility-based features, forming the input to a gradient-boosted decision tree classifier (XGBoost). To ensure interpretability, we employ SHAP (SHapley Additive exPlanations), which decomposes each prediction into additive feature attributions while satisfying game-theoretic fairness axioms. In addition, we integrate differential privacy into the training pipeline to ensure robustness against membership inference attacks and protect proprietary or client-sensitive data. Empirical evaluations across multiple S&P 500 equities from 2018–2023 demonstrate that our FinBERT-enhanced model consistently outperforms both technical-only and lexicon-based sentiment baselines in terms of AUC, F1-score, and simulated trading profitability. SHAP analysis confirms that FinBERT-derived features rank among the most influential predictors. Our findings highlight the complementary value of domain-specific NLP and privacy-preserving machine learning in financial forecasting, offering a principled, interpretable, and deployable solution for real-world quantitative finance applications. Full article

Energy poverty is a multidimensional phenomenon that impairs access to basic energy services and threatens social well-being, particularly in disadvantaged rural communities. This study investigates the extent and drivers of household energy poverty in a Hungarian village through a survey-based analysis (N = 257) conducted in early 2025. The sample is not nationally representative, however, it reflects approximately 20% of the total village population (1331 inhabitants). This study aims to identify vulnerable household profiles, explore correlations between socio-economic and housing factors and perceived thermal comfort, and compare the effectiveness of multiple measurement indicators the 10% rule, low income high cost, 2M, and M/2. We employ descriptive statistics, Pearson correlation, Fuzzy C-Means clustering, and linear regression, revealing that over half of the sample is energy poor according to the 10% rule, while the LIHC method identifies 29%. Our regression results confirm that cluster membership significantly influences perceived comfort levels (R² = 0.063, p = 0.002). We conclude that single-indicator approaches are insufficient to capture the nuanced realities of rural energy poverty, therefore, we recommend the development of a rural energy poverty index. Such a tool could help identify affected households and support the formulation of context-sensitive, evidence-based energy and social policy interventions. Full article

Simulated data created in silico using a previously reported method were sampled by bootstrapping to generate data sets for training multiple copies of an ensemble learner (i.e., a machine learning (ML) method). The posterior probabilities of class membership obtained by applying the ensemble of ML models to previously unseen validation data were fitted to a beta distribution. The shape parameters for the fitted distribution were used to calculate the subjective opinion of sample membership into one of two mutually exclusive classes. The subjective opinion consists of belief, disbelief and uncertainty masses. A subjective opinion for each validation sample allows identification of high-uncertainty predictions. The projected probabilities of the validation opinions were used to calculate log-likelihood ratio scores and generate receiver operating characteristic (ROC) curves from which an opinion-supported decision can be made. Three very different ML models, linear discriminant analysis (LDA), random forest (RF), and support vector machines (SVM) were applied to the two-state classification problem in the analysis of forensic fire debris samples. For each ML method, a set of 100 ML models was trained on data sets bootstrapped from 60,000 in silico samples. The impact of training data set size on opinion uncertainty and ROC area under the curve (AUC) were studied. The median uncertainty for the validation data was smallest for LDA ML and largest for the SVM ML. The median uncertainty continually decreased as the size of the training data set increased for all ML.The AUC for ROC curves based on projected probabilities was largest for the RF model and smallest for the LDA method. The ROC AUC was statistically unchanged for LDA at training data sets exceeding 200 samples; however, the AUC increased with increasing sample size for the RF and SVM methods. The SVM method, the slowest to train, was limited to a maximum of 20,000 training samples. All three ML methods showed increasing performance when the validation data was limited to higher ignitable liquid contributions. An ensemble of 100 RF ML models, each trained on 60,000 in silico samples, performed the best with a median uncertainty of 1.39x${10}^{-2}$ and ROC AUC of 0.849 for all validation samples. Full article

Rough sets and fuzzy sets are two complementary approaches for modeling uncertainty and imprecision. Their integration enables a more comprehensive representation of complex, uncertain systems. However, existing rough fuzzy sets models lack the expressive power to fully capture the interactions among structural uncertainty, cognitive hesitation, and multi-level granular information. To address these limitations, we achieve the following: (1) We propose intuitionistic fuzzy covering rough membership and non-membership degrees based on maximal description and construct a new single-granulation model that more effectively captures both the structural relationships among elements and the semantics of fuzzy information. (2) We further extend the model to a multi-granulation framework by defining optimistic and pessimistic approximation operators and analyzing their properties. Additionally, we propose a neutral multi-granulation covering rough intuitionistic fuzzy sets based on aggregated membership and non-membership degrees. Compared with single-granulation models, the multi-granulation models integrate multiple levels of information, allowing for more fine-grained and robust representations of uncertainty. Finally, a case study on real estate investment was conducted to validate the effectiveness of the proposed models. The results show that our models can more precisely represent uncertainty and granularity in complex data, providing a flexible tool for knowledge representation in decision-making scenarios. Full article

With the increasing penetration of distributed generation (DG), integrated main-distribution networks (IMDNs) face challenges in rapidly and effectively performing comprehensive operational risk assessments under multiple uncertainties. Thereby, using the traditional hierarchical economic scheduling method makes it difficult to accurately find the optimal scheduling operating point. To address this problem, this paper proposes a multi-objective dispatch decision-making optimization model for the IMDN with static security region (SSR) constraints. Firstly, the non-sequential Monte Carlo sampling is employed to generate diverse operational scenarios, and then the key risk characteristics are extracted to construct the risk assessment index system for the transmission and distribution grid, respectively. Secondly, a hyperplane model of the SSR is developed for the IMDN based on alternating current power flow equations and line current constraints. Thirdly, a risk assessment matrix is constructed through optimal power flow calculations across multiple load levels, with the index weights determined via principal component analysis (PCA). Subsequently, a scheduling optimization model is formulated to minimize both the system generation costs and the comprehensive risk, where the adaptive grid density-improved multi-objective particle swarm optimization (AG-MOPSO) algorithm is employed to efficiently generate Pareto-optimal operating point solutions. A membership matrix of the solution set is then established using fuzzy comprehensive evaluation to identify the optimal compromised operating point for dispatch decision support. Finally, the effectiveness and superiority of the proposed method are validated using an integrated IEEE 9-bus and IEEE 33-bus test system. Full article

Background/Objectives: Atypical sensory processing is increasingly recognized as a transdiagnostic dimension of neurodevelopmental disorders (NDDs), with critical implications for emotional and behavioral regulation. This study aimed to identify distinct sensory profiles in preschool children with NDDs and to examine their associations with emotional–behavioral and cognitive/developmental functioning. Methods: A total of 263 children (aged 21–71 months) diagnosed with autism spectrum disorder (ASD), language disorder (LD), or other NDDs (ONDD) were recruited. Sensory processing was assessed using the SPM-P, emotional–behavioral functioning was assessed via the CBCL 1½–5, and cognitive/developmental levels were assessed through standardized instruments. Latent profile analysis (LPA) was conducted to identify sensory subtypes. Group comparisons and multinomial logistic regression were used to examine profile characteristics and predictors of profile membership. Results: Three sensory profiles emerged: (1) Multisystemic Sensory Dysfunction (20.1%), characterized by pervasive sensory and emotional difficulties, primarily observed in ASD; (2) Typical Sensory Processing (44.9%), showing normative sensory and emotional functioning, predominantly LD; and (3) Mixed Subclinical Sensory Processing (35%), with subclinical-range scores across multiple sensory and emotional domains, spanning all diagnoses. Higher cognitive functioning and fewer internalizing symptoms significantly predicted membership in the typical profile. A gradient of symptom severity was observed across profiles, with the Multisystemic group showing the most pronounced emotional–behavioral impairments. Conclusions: Distinct sensory–emotional phenotypes were identified across diagnostic categories, supporting a dimensional model of neurodevelopment. Sensory profiles were strongly associated with emotional functioning, independently of diagnostic status. Early sensory assessment may therefore offer clinically meaningful insights into emotional vulnerability and inform targeted interventions in preschool populations with NDDs. Full article

Meteorological observations are essential for climate modelling, prediction, early warning systems, decision-making processes, and disaster management. These observations are critical to societal development and the safeguarding of human activities and livelihoods. Spatial interpolation techniques play a pivotal role in addressing gaps between observation sites, enabling the generation of continuous meteorological datasets. However, due to the inherent complexity of atmosphere–surface interactions, no single interpolation technique has proven universally effective in achieving consistently accurate results for meteorological variables. This study proposes a novel interpolation model based on Fuzzy Adaptive Optimal Fusion (FAOF). The FAOF model integrates fuzzy theory by constructing station-specific fuzzy sets and sub-method element pools, employing a nonlinear membership function with error as the independent variable. An iterative accuracy index is used to identify the optimal parameter combination, facilitating adaptive data fusion and interpolation optimisation. The model’s performance is evaluated against 10 individual methods from the method pool. Experimental results demonstrate that FAOF effectively combines the strengths of multiple methods, achieving significantly enhanced interpolation accuracy. Additionally, the model consistently performs well across diverse regions and meteorological variables, underscoring its robustness and strong generalisation capability. Full article

One of the most challenging problems in data analysis is visualizing patterns and extracting insights from multi-dimensional datasets that vary over time. The complexity of data and variations in the correlations between different features adds further difficulty to the analysis. In this paper, we provide a framework to analyze the temporal dynamics of such datasets. We use machine learning clustering techniques and examine the time evolution of data patterns by constructing the corresponding cluster trajectories. These trajectories allow us to visualize the patterns and the changing nature of correlations over time. The similarity and correlations of features are reflected in common cluster membership, whereas the historical dynamics are described by a trajectory in the corresponding (cluster, time) space. This allows an effective visualization of multi-dimensional data over time. We introduce several statistical metrics to measure duration, volatility, and inertia of changes in patterns. Using the Hamming distance of trajectories over multiple time periods, we propose a novel metric, the Hamming diversification index, to measure the spread between trajectories. The novel metric is easy to compute, has a simple machine learning implementation, and provides additional insights into the temporal dynamics of data. This parsimonious diversification index can be used to examine changes in pattern similarities over aggregated time periods. We demonstrate the efficacy of our approach by analyzing a complex multi-year dataset of multiple worldwide economic indicators. Full article