Expert Systems最新文献

{"title":"FLAMES—Federated Learning for Advanced MEdical Segmentation","authors":"Martina Savoia,&nbsp;Edoardo Prezioso,&nbsp;Francesco Piccialli","doi":"10.1111/exsy.70090","DOIUrl":"https://doi.org/10.1111/exsy.70090","url":null,"abstract":"<p>Federated learning (FL) is gaining traction across numerous fields for its ability to foster collaboration among multiple participants while preserving data privacy. In the medical domain, FL enables institutions to share knowledge while maintaining control over their data, which often vary in modality, source, and quantity. Institutions are often specialised in treating one or a few types of tumours, typically focusing on a specific organ. Hence, different institutions may contribute with distinct types of medical imaging data of various organs, originating from diverse machines. Collaboration among these institutions enhances performance on shared tasks across different areas of the body. The framework employs modality-specific models hosted on the server, each designed for a particular imaging modality and designed to predict the presence of tumours in scans from its respective modality, regardless of the organ being imaged. Clients focus on their specific imaging modality, utilising knowledge derived from images contributed by institutions employing the same modality. This approach facilitates broader collaboration, extending beyond institutions specialising in the same organ to include those working within the same imaging modality. This approach also helps avoid the introduction of potential noise from clients with images of different modalities, which might hinder the model's ability to effectively specialise and adapt to the data specific to each institution. Experiments showed that FLAMES achieves strong performance on server data, even when tested across different organs, demonstrating its ability to generalise effectively across diverse medical imaging datasets. Our code is available at https://github.com/MODAL-UNINA/FLAMES.</p>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"42 8","pages":""},"PeriodicalIF":3.0,"publicationDate":"2025-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1111/exsy.70090","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144472904","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Federated Cross-Domain Recommendation Framework With Graph Neural Network","authors":"Deling Huang,&nbsp;Qilong Feng","doi":"10.1111/exsy.70087","DOIUrl":"https://doi.org/10.1111/exsy.70087","url":null,"abstract":"<div>\u0000 \u0000 <p>Cross-domain recommendation (CDR) leverages more abundant source-domain information to improve target-domain recommendation accuracy. However, traditional centralized CDR approaches face two critical limitations: (1) centralized data storage causes privacy vulnerabilities against malicious servers, and (2) gradient leakage during uploading enables recovery of source data. To address these challenges, in this work, we propose FedGraphCDR, a federated learning-based cross-domain recommendation framework that integrates local differential privacy (LDP) with pseudo item injection during gradient aggregation to prevent gradient leakage attacks, while utilizing graph neural networks to identify comparable users and mitigate cold-start problems. Evaluation on a real-life Douban dataset spanning three domains demonstrates that our framework successfully combines LDP with pseudo items to enhance privacy protection while achieving superior recommendation accuracy over benchmark methods. The results confirm that FedGraphCDR effectively resolves privacy concerns and improves recommendation quality, particularly for cold-start users, and establishes a practical solution for privacy-preserving cross-domain recommendation.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"42 8","pages":""},"PeriodicalIF":3.0,"publicationDate":"2025-06-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144339607","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Hybrid Spiking Model for Anomaly Detection in Multivariate Time Series","authors":"Wei Zhang,&nbsp;Ping He,&nbsp;Shengrui Wang,&nbsp;Fan Yang,&nbsp;Ying Liu","doi":"10.1111/exsy.70086","DOIUrl":"https://doi.org/10.1111/exsy.70086","url":null,"abstract":"<div>\u0000 \u0000 <p>Deep neural networks have exhibited preeminent performance in anomaly detection, but they struggle to effectively capture changes over time in multivariate time-series data and suffer from resource consumption issues. Spiking neural networks address these limitations by capturing the change in time-varying signals and decreasing resource consumption, but they sacrifice performance. This paper develops a novel spiking-based hybrid model incorporated a temporal prediction network and a reconstruction network. It integrates a unique first-spike frequency encoding scheme and a firing rate based anomaly score method. The encoding scheme enhances the event representation ability, while the anomaly score enables efficient anomaly identification. Our proposed model not only maintains low resource consumption but also improves the ability of anomaly detection. Experiments on publicly real-world datasets confirmed that the proposed model acquires state-of-the-art performance superior to existing approaches. Remarkably, it costs 5.04× lower energy consumption compared with the artificial neural network version.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"42 8","pages":""},"PeriodicalIF":3.0,"publicationDate":"2025-06-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144323439","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enabling the Application of Graph Neural Networks on Graphs With Unknown Connectivity","authors":"Jorge García-Carrasco,&nbsp;Alejandro Maté,&nbsp;Juan Trujillo","doi":"10.1111/exsy.70088","DOIUrl":"https://doi.org/10.1111/exsy.70088","url":null,"abstract":"<p>Graph Neural Networks (GNNs) have proven to be reliable methods for working with graph-structured data. However, it is common to find graphs with partially or fully inaccessible connectivity patterns, hindering the direct application of GNNs to the task at hand. To tackle this problem, several Graph Structure Learning (GSL) methods have been proposed, with the objective of jointly optimizing both the graph structure and the GNN model by adding loss terms that enforce desired graph properties. These properties, such as sparseness and connectivity of similar nodes, can have a drastic impact on the performance of a GNN. However, current methods offer little control on the desired degree of sparseness, which may lead to non-optimal connectivity and reduced efficiency. In this paper, we propose a new method called Adaptative Sparsification Graph Learning (ASGL), which enables fine-grained, linear control over the total number of edges in the resulting learned graph via a novel perturbation-based loss term. ASGL not only provides flexibility in sparsity control but also improves both accuracy and computational efficiency, outperforming state-of-the-art methods in most benchmarks. We demonstrate its robustness through extensive experiments and highlight how adjusting sparsity enables optimizing the trade-off between accuracy, complexity, and interpretability.</p>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"42 8","pages":""},"PeriodicalIF":3.0,"publicationDate":"2025-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1111/exsy.70088","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144315183","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Toxic Discourse in the Digital Battlefield: Analysing Telegram Channels During the Russia–Ukraine ‘Conflict’","authors":"Arsenii Tretiakov,&nbsp;Sergio D'Antonio-Maceiras,&nbsp;Áurea Anguera de Sojo Hernández,&nbsp;Alejandro Martín","doi":"10.1111/exsy.70081","DOIUrl":"https://doi.org/10.1111/exsy.70081","url":null,"abstract":"<p>Instant messenger Telegram has emerged as a favoured platform for far-right activism, conspiracy theories, political propaganda, and misinformation, which has its own target audience. This study explores the application of multilingual pre-trained language models to detect and measure toxicity in political content on Telegram channels. The proposed techniques have shown notable advancements in identifying toxic information using a fine-tuned RoBERTa model. Through the combination of data analysis, time-series analysis, and BERTopic modelling, the research demonstrates how toxicity varies by topic, country, and time period, using metadata. The study identified key topics in the dataset, which includes 23.6 million messages from 1491 Telegram channels, including the Russian–Ukrainian conflict and political tensions in Europe and the United States from 2016 to 1 July 2023. Despite these achievements, challenges such as the dominance of Russian language content and a focus on specific topics were highlighted. This research advances the understanding of how toxic language and propaganda are disseminated across different languages and political narratives, contributing to the study of digital communication and information warfare.</p>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"42 7","pages":""},"PeriodicalIF":3.0,"publicationDate":"2025-06-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1111/exsy.70081","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144292778","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Uncertainty-Guided Diffusion Model for High-Fidelity ECG Synthesis and Classification","authors":"Qi Zhang,&nbsp;Hongyan Li","doi":"10.1111/exsy.70070","DOIUrl":"https://doi.org/10.1111/exsy.70070","url":null,"abstract":"<div>\u0000 \u0000 <p>Electrocardiogram (ECG) synthesis plays a crucial role in medical research, education and device development. However, achieving high-fidelity ECG signal synthesis remains challenging, particularly in accurately reproducing specific waveform patterns at the sample level. In this paper, we propose an uncertainty-guided diffusion model that integrates uncertainty estimation into the ECG synthesis process. The uncertainty guidance preserves meaningful waveform characteristics. The model combines diffusion models, known for generating high-quality samples from complex distributions, with uncertainty guidance that captures and propagates uncertainty throughout the pipeline. Extensive experiments demonstrate that our approach outperforms existing methods in terms of both distribution-level and sample-level evaluation.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"42 7","pages":""},"PeriodicalIF":3.0,"publicationDate":"2025-06-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144292969","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Single and Ensemble Based Filters in Environmental Data","authors":"Yousra Cherif,&nbsp;Ali Idri","doi":"10.1111/exsy.70076","DOIUrl":"https://doi.org/10.1111/exsy.70076","url":null,"abstract":"<div>\u0000 \u0000 <p>Researchers rely on species distribution models (SDMs) to establish a correlation between species occurrence records and environmental data. These models offer insights into the ecological and evolutionary aspects of the subject. Feature selection (FS) aims to choose useful interlinked features or remove unnecessary and redundant ones and make the induced model easier to understand. Although feature selection plays a crucial role in SDMs, only a limited number of studies in the literature have addressed it with several key shortcomings such as lack of the use of multivariate techniques, lack of comparison between the univariate and the multivariate filters, and absence of a comparison between the ensemble univariate and multivariate filters. Therefore, this study presents a rigorous empirical evaluation consisting of assessing and comparing six filter-based univariate feature selection methods using two thresholds with two multivariate techniques, as well as four classifiers: Extreme Gradient boosting (XGB), Random Forest (RF), Decision Tree (DT), and Light gradient-boosting machine (LGBM). Furthermore, the current study proposes a novel approach for ensemble construction consisting of evaluating the applications of ensemble learning using 40% of features ranked by means of Borda Count and Reciprocal Rank (univariate filter ensembles) as well as the fusion-based and the intersection-based ensembles (multivariate filter ensembles). Moreover, we evaluated and compared the performances of univariate and multivariate techniques with their ensembles. Similarly, we evaluated and compared the performances of the best ensemble techniques across datasets. The empirical evaluations involve several techniques, such as the 5-fold cross-validation method, the Scott Knott (SK) test, and Borda Count. In addition, we used three performance metrics (accuracy, Kappa, and <i>F</i>1-score). Experiments showed that Consistency-based subset selection in conjunction with RF outperformed all other univariate and multivariate FS techniques with an accuracy value of 91.63% across all datasets. However, Fisher score trained with RF was the best choice when considering the number of features. Moreover, the univariate or multivariate based ensembles, in general, outperformed their singles. In addition, when comparing the univariate and multivariate ensembles, the fusion-based ensemble outperformed all other ensembles achieving an accuracy of 91.77% when using RF across datasets. Nevertheless, in terms of performance and number of features, the ensemble constructed using Reciprocal Rank performed better than all other FS techniques regardless of the classifier used. It achieved an accuracy of 91.61% across datasets when using RF.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"42 7","pages":""},"PeriodicalIF":3.0,"publicationDate":"2025-06-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144273531","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Transforming Earth Observation: An Extensive Evaluation of Vision Transformers for Satellite Images-Based Land Cover Classification","authors":"Fakhri Alam Khan","doi":"10.1111/exsy.70082","DOIUrl":"https://doi.org/10.1111/exsy.70082","url":null,"abstract":"<div>\u0000 \u0000 <p>Satellite imagery offers rich information for land cover classification, but choosing an effective yet efficient feature extractor or backbone architecture remains challenging. In this study, I benchmark 25 vision-transformers across 10 public land cover datasets to guide backbone selection for downstream classification tasks. The proposed approach encodes each satellite image into a fixed-length feature vector via a pre-trained transformer, then trains and tests a linear support-vector classifier on these encodings to isolate the impact of the backbone alone. I report average classification accuracy and F1-score over three random stratified splits per dataset, and I also measure training time to assess the computational cost. Results show that the image encoding performed using large-receptive-field transformers with advanced self-attention—particularly <span>deit3_base_patch16_224</span> and <span>twins_svt_large</span>—achieve the highest accuracies without incurring prohibitive training times. In contrast, encodings of the compact variants achieve faster training but incur notable performance drops around 7%–8%. These findings reveal a clear trade-off between representational power and efficiency. Practitioners can leverage such rankings to select a transformer backbone that best balances accuracy and computational efficiency for satellite image-based land cover classification tasks, accelerating the development of robust and resource-aware systems.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"42 7","pages":""},"PeriodicalIF":3.0,"publicationDate":"2025-06-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144256076","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Method for Extracting Black Box Models Based on Interpretable Attention","authors":"Lijun Gao,&nbsp;Huibin Tian,&nbsp;Kai Liu","doi":"10.1111/exsy.70084","DOIUrl":"https://doi.org/10.1111/exsy.70084","url":null,"abstract":"<div>\u0000 \u0000 <p>Deep neural networks have achieved remarkable success in face recognition. However, their vulnerability has attracted considerable attention. Researchers can analyse the weaknesses of face recognition models by extracting their functionality, aiming to enhance the security performance of these models. The findings of the study reveal that current model extraction methods are afflicted with notable drawbacks, namely low similarity in capturing model functionality and insufficient availability of samples. These limitations significantly impede the analysis of model security performance. We propose an interpretable attention-based method for black-box model extraction, enhancing the similarity between substitute and victim model functionality. Our main contributions are summarized as follows: (i) This study addresses the issue of limited sample training caused by the restricted number of black-box hard label queries. (ii) By applying input perturbations, we obtain feedback from deep black-box models, enabling us to identify facial local regions and the distribution of feature weights that positively influence predictions. (iii) By normalizing the feature weight distribution matrix and associating it with the attention weight matrix, the construction of an attention mask for the dataset is achieved, enabling differential attention to features in different regions. (iv) Leveraging a pre-trained base model, we extract relevant knowledge and features, facilitating cross-domain knowledge transfer. Experiments on Emore, PubFig and CASIA-WebFace show that our method outperforms traditional methods by 10%–20% in model consistency for the same query budget. Also, our method achieves the highest model stealing consistency on the three datasets: 94.51%, 93.27% and 91.74%, respectively.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"42 7","pages":""},"PeriodicalIF":3.0,"publicationDate":"2025-06-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144206835","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Optimising Performance Curves for Ensemble Models through Pareto Front Analysis of the Decision Space","authors":"Alberto Gutierrez-Gallego,&nbsp;Oscar Garnica,&nbsp;Daniel Parra,&nbsp;J. Manuel Velasco,&nbsp;J. Ignacio Hidalgo","doi":"10.1111/exsy.70075","DOIUrl":"https://doi.org/10.1111/exsy.70075","url":null,"abstract":"<p>Receiver operating characteristic curves are commonly used to evaluate the performance of machine learning ensemble classification models that combine multiple classifiers through a voting procedure. Although these models have many parameters, standard ROC analyses typically vary only the voting threshold, limiting their potential for improvement. In this paper, we propose <span>Performance Curve Mapping</span>, a new method that redefines the ROC curve as the Pareto front of a multi-objective optimisation problem. The method maps the multidimensional space of all ensemble parameters (Decision space) into a two-dimensional Objective space defined by classification performance metrics. We employ an algorithm based on NSGA-II to explore the Decision space and validate the proposal on two different classification problems: (1) predicting car insurance claims in a highly imbalanced dataset (<span>Insurance</span> dataset), and (2) predicting obesity risk in a balanced clinical dataset (<span>GenObIA</span> dataset). We compare our method with alternative ensemble optimisation approaches, using visual assessment, the area under the curve and the Youden index as performance measures. In the <span>Insurance</span> dataset, <span>Performance Curve Mapping</span> achieves an average improvement of 46.4% in AUC-ROC and 26.1% in the Youden index. In the <span>GenObIA</span> dataset, it achieves an average improvement of 29.7% in AUC-ROC and 11.9% in the Youden index. All improvements are calculated relative to the maximum achievable improvement.</p>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"42 7","pages":""},"PeriodicalIF":3.0,"publicationDate":"2025-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1111/exsy.70075","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144171891","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}