Expert Systems最新文献

{"title":"A Stochastic Agnostic Meta-Clustering Framework: Application to African Agro-Ecological Zoning","authors":"Naila Aziza Houacine,&nbsp;Adel Got,&nbsp;Widad Hassina Belkadi,&nbsp;Habiba Drias,&nbsp;Rached Hamidi,&nbsp;Mohamed Rafik Douha","doi":"10.1111/exsy.70438","DOIUrl":"https://doi.org/10.1111/exsy.70438","url":null,"abstract":"<div>\u0000 \u0000 <p>This article introduces Stochastic Agnostic Meta-Clustering (SAMC), a novel ensemble clustering framework designed to address the challenges of high-dimensional and uncertain data. SAMC integrates stochastic feature subspace decomposition with metaheuristic optimization to support both hard and soft clustering, making it well suited for domains characterized by gradual transitions and fuzzy boundaries, such as environmental and agricultural systems. The proposed framework is evaluated on an agro-ecological zoning (AEZ) task using data structured around SCORPAN factors and is compared against established hard and soft clustering methods, with and without dimensionality reduction. Experimental results demonstrate the SAMC's advantageous trade-off between interpretability and scalability while improving robustness to feature heterogeneity and preventing over-fragmentation. In addition, post-clustering interpretability analyses are conducted to identify dominant agro-ecological drivers and representative rules characterizing the resulting zones. The comparative analysis further confirms that soft clustering approaches are better suited to capturing the fuzzy boundaries inherent in environmental systems and provided a soft clustering-based AEZ analysis at the continental scale for Africa.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"43 11","pages":""},"PeriodicalIF":3.4,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149146966","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":"Group Emotion Recognition Using Hybrid-Level Fusion Strategy","authors":"Jingfeng Deng,&nbsp;Xinxin Wu,&nbsp;Ziwei Cui,&nbsp;Xingzhi Wang","doi":"10.1111/exsy.70435","DOIUrl":"https://doi.org/10.1111/exsy.70435","url":null,"abstract":"<div>\u0000 \u0000 <p>Group Emotion Recognition (GER) plays a vital role in public safety by enabling proactive risk warnings and resource allocation. However, current approaches often rely solely on visual emotional cues, and their performance is limited by linear fusion mechanisms. To overcome these limitations, this paper introduces a novel multimodal fusion framework that leverages Kolmogorov-Arnold Networks (KANs) to effectively integrate multisource information. Our approach begins by extracting emotional features from three distinct modalities: group image scenes are processed by ResNet-50, facial regions are analysed using VGGFace and textual descriptions, generated by the Gemini 2.0 model, are encoded with ELECTRA-base. These features are then fused through an innovative hybrid-level strategy based on KAN, which leverages its superior nonlinear modelling capability to generate robust emotional representations. Extensive experiments on three benchmark datasets, GAF 2.0, GAF 3.0 and GroupEmoW, demonstrate the consistent superiority of our method, which achieves state-of-the-art accuracies of 92.87%, 92.01% and 91.88%, respectively. The advantage of the KAN-based fusion is further validated through a comparison against seven alternative fusion mechanisms, paired statistical significance tests and an interpretability analysis of the learned modality contributions; moreover, we show that an open-source MLLM (Qwen3-VL) can replace the proprietary model with only a modest accuracy trade-off, improving reproducibility and deployability.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"43 11","pages":""},"PeriodicalIF":3.4,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149146986","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 Comprehensive Study of Graph Neural Networks: Types and Application","authors":"Rahul Dubey,&nbsp;Ramandeep Kaur,&nbsp;Nidhi Ahuja","doi":"10.1111/exsy.70429","DOIUrl":"https://doi.org/10.1111/exsy.70429","url":null,"abstract":"<div>\u0000 \u0000 <p>This systematic review presents a comprehensive analysis of graph neural networks (GNNs), focusing on their major types and applications. GNNs have emerged as an effective deep learning (DL) method for modelling and analysing intricate relationships in graph-structured data. Their versatility has led to widespread adoption across diverse domains and tasks. This review examines a wide range of GNN models and their applications, analysing their strengths and limitations. A total of 96 research papers published between 2019 and 2025 are systematically reviewed, providing a comprehensive overview of the development and application of GNN-based methods. Despite the remarkable success of GNNs across diverse application domains, existing models still face several challenges, including scalability, high computational complexity, limited interpretability, and inadequate generalisation to heterogeneous and evolving data environments. Future research should focus on developing lightweight, explainable, and privacy-preserving GNN architectures, integrating self-supervised and multimodal learning strategies, and designing robust frameworks capable of handling dynamic and large-scale graph data. This review provides a valuable reference for practitioners and researchers in the field of GNNs, supporting future research and application development.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"43 11","pages":""},"PeriodicalIF":3.4,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149146929","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":"Forecasting Deployment-Time Blind Spots in Language-Model Intrusion Detection Agents: A Data-Geometric Coverage Statistic and the Relay Condition That Carries It to the Agent","authors":"Qianli Di,&nbsp;Keju Du,&nbsp;Gang Chen,&nbsp;Hai Deng","doi":"10.1111/exsy.70436","DOIUrl":"https://doi.org/10.1111/exsy.70436","url":null,"abstract":"<div>\u0000 \u0000 <p>An operator deploying an intrusion detector on a new attack surface cannot say which attack families it will silently miss. For tree-ensemble detectors, that can be forecast before deployment from labelled traffic, with confirmation on one of the two benchmarks we test. For a candidate family, a nearest-neighbour coverage statistic computed from labelled known traffic and a small sample of that family uses no model output and nothing about the detector. On the 11 CIC-IDS2017 leave-one-attack-out folds the forecast survives rank correlation, bootstrap confidence intervals and multiplicity correction for both tree-ensemble families (random forest <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>r</mi>\u0000 <mo>=</mo>\u0000 <mn>0.90</mn>\u0000 </mrow>\u0000 <annotation>$$ r=0.90 $$</annotation>\u0000 </semantics></math>, 0.87 across reference-set draws, CI [0.68, 0.99]); on UNSW-NB15 it holds directionally only. Ten labelled samples suffice (<span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mi>r</mi>\u0000 <mo>=</mo>\u0000 <mn>0.82</mn>\u0000 </mrow>\u0000 <annotation>$$ r=0.82 $$</annotation>\u0000 </semantics></math> against <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mn>0.86</mn>\u0000 </mrow>\u0000 <annotation>$$ 0.86 $$</annotation>\u0000 </semantics></math> at 200). Reaching the deployed language-model agent is what harness engineering contributes: SecHarness, a five-component execution-layer specification, constrains a 3B-parameter agent into a relay whose blind spots coincide with its tool's. We report the null in full: against the same random-forest verdict written into the prompt, the harness changes 1 decision in 200 at <span></span><math>\u0000 <semantics>\u0000 <mrow>\u0000 <mn>1.73</mn>\u0000 <mo>×</mo>\u0000 </mrow>\u0000 <annotation>$$ 1.73times $$</annotation>\u0000 </semantics></math> latency, and fine-tuning adds neither accuracy nor speed on matched hardware, while the fine-tuned 3B agent fabricates tool citations once its tool is withdrawn. It keeps rather than creates what the tool supplies: catching every error costs 40.5% of the alert stream under the harness against the detector's own 43.5%, a six-record gap we report as an artefact. Knowing where an agent will be blind should precede optimising the model inside.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"43 11","pages":""},"PeriodicalIF":3.4,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149146859","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":"Mission-Aligned Learning-Informed Control of Autonomous Systems: Formulation and Foundations","authors":"Vyacheslav Kungurtsev,&nbsp;Alessandro Di Frenna,&nbsp;Gustav Šír,&nbsp;Monicah Cherop Naibei,&nbsp;Haozhe Tian,&nbsp;Homayoun Hamedmoghadam,&nbsp;Akhil Anand,&nbsp;Sebastien Gros","doi":"10.1111/exsy.70433","DOIUrl":"https://doi.org/10.1111/exsy.70433","url":null,"abstract":"<p>Research, innovation and practical capital investment have been increasing rapidly toward the realization of autonomous physical agents. This includes industrial and service robots, unmanned aerial vehicles, embedded control devices, and a number of other realizations of cybernetic/mechatronic implementations of intelligent autonomous devices. In this paper, we consider a stylized version of robotic care, which would normally involve a two-level Reinforcement Learning procedure that trains a policy for both lower level physical movement decisions as well as higher level conceptual tasks and their sub-components. In order to deliver greater safety and reliability in the system, we present the general formulation of this as a two-level expert defined optimization scheme which incorporates control with physics modelling of the system and environment at the lower level, and classical planning with complex task logic at the higher level, integrated with a capacity for learning. This synergistic integration of multiple methodologies—control, classical planning, and RL—presents an opportunity for greater insight for algorithm development, leading to more efficient and reliable performance. Here, the notion of reliability pertains to physical safety and interpretability into an otherwise black box operation of autonomous agents, concerning users and regulators. This work presents the principled integration of the two expert-defined domains. The methodology is validated in a simulation of robotic care aid, with complex medication and meal preparation tasks at the upper level and locomotion at the lower level. The validation confirms the framework's adaptability to human preferences and symbolic to continuous feedback.</p>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"43 11","pages":""},"PeriodicalIF":3.4,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1111/exsy.70433","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149146962","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":"Large Language Models and Agentic AI for Vulnerability Management: A Systematic Review","authors":"Yasamin Akrami,&nbsp;Malek Malkawi,&nbsp;Reda Alhajj","doi":"10.1111/exsy.70437","DOIUrl":"https://doi.org/10.1111/exsy.70437","url":null,"abstract":"<p>Large Language Models (LLMs), pretrained language and code models and agentic artificial intelligence are increasingly being investigated across the vulnerability-management lifecycle. However, the evidence remains fragmented across code-level prediction, vulnerability-intelligence enrichment, exploit-oriented evaluation, automated repair and bounded defensive action. Following PRISMA 2020 guidance, this systematic review searched six scholarly databases, supplemented the initial search with an update search conducted on 22 August 2026 and backward and forward citation searching, and identified 53 eligible peer-reviewed primary studies published between 2023 and 2026. The studies were organized into four application areas: vulnerability discovery and detection; vulnerability intelligence and prioritization; threat context, exploitation and operational hunting; and remediation, response and autonomous defence. Model-centred studies provide the most standardized benchmark evidence, but their reported performance remains sensitive to dataset construction, class imbalance, leakage, code context, language coverage and cross-dataset transfer. More recent work increasingly combines retrieval, structured program or security knowledge, external tools, multi-agent coordination and verification; however, only 15 of the 53 studies (28.3%) implemented or directly evaluated systems meeting this review's operational definition of genuinely agentic behaviour through adaptive evidence acquisition, iterative tool use, feedback-driven control, coordination, or bounded action. Exploit-oriented and autonomous-defence evaluations provide useful capability evidence but should not be interpreted as direct evidence of operational vulnerability-management autonomy. Across remediation and response studies, execution success alone does not establish vulnerability removal, semantic correctness, operational safety, or production readiness. To separate technical results from operational credibility, the review applies an evidence-maturity framework and analyses recurring forms of data, evaluation, grounding, verification and governance debt. The findings support bounded and auditable use of LLMs within evidence-grounded vulnerability-management workflows rather than unrestricted autonomous operation.</p>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"43 11","pages":""},"PeriodicalIF":3.4,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1111/exsy.70437","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149146860","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":"Data-Driven Vertiport Site Selection: A Two-Stage Framework With Multi-Source Urban Data and Scalable Optimization","authors":"Zhirong Zhang,&nbsp;Yucheng Wang,&nbsp;Yuan Liang,&nbsp;Qixiu Cheng,&nbsp;Yong Fang","doi":"10.1111/exsy.70423","DOIUrl":"https://doi.org/10.1111/exsy.70423","url":null,"abstract":"<div>\u0000 \u0000 <p>Urban Air Mobility (UAM) is increasingly recognized as a promising solution to alleviate urban congestion and improve transportation efficiency, playing a key role in developing sustainable urban infrastructure. A central challenge for UAM deployment is the siting of vertiports, requiring the balancing of various urban factors such as demand, traffic, housing prices, and noise. Existing methods rely on heuristic or simplified optimization and are tailored to individual cities, resulting in poor transferability and limited scalability. This paper introduces a novel two-stage framework for UAM vertiport location selection. In the first stage, we develop DRAG-Net, a Dual-stream Recency-biased Attention Graph Network, to learn transferable candidate-site representations from multi-source urban data and identify potential vertiport locations. In the second stage, we apply the non-dominated sorting genetic Algorithm II (NSGA-II) to generate Pareto-optimal layouts from these candidate sites, maximizing demand coverage while minimizing noise. To evaluate effectiveness and generalization, we construct comprehensive datasets for Beijing and Shanghai that integrate geographical, socio-economic, and temporal features. The results show that the model trained in Beijing and effectively transferred to Shanghai preserves the learned spatial structure and achieves an effective multi-objective trade-off between demand and noise, thus supporting its cross-city generalizability. Furthermore, we enhance interpretability by employing SHAP analysis, which provides model-agnostic explanations and transparent insights into the decision process. Overall, this study establishes a scalable and generalizable framework for UAM infrastructure planning by combining predictive modelling with multi-objective optimization and interpretable analysis.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"43 11","pages":""},"PeriodicalIF":3.4,"publicationDate":"2026-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149048567","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 Semi-Interpretable Multimodal Model Combining Electronic Health Records and Chest X-Rays for COVID-19-Related Death Prediction","authors":"Lander Rodriguez-Idiazabal,&nbsp;Jose M. Quintana,&nbsp;Javier Del Ser,&nbsp;Jose Javier Echevarria-Uraga,&nbsp;Maria Jose Legarreta,&nbsp;Nere Larrea,&nbsp;Fernando García-García,&nbsp;Irantzu Barrio","doi":"10.1111/exsy.70440","DOIUrl":"https://doi.org/10.1111/exsy.70440","url":null,"abstract":"&lt;div&gt;\u0000 \u0000 \u0000 &lt;section&gt;\u0000 \u0000 &lt;h3&gt; Purpose&lt;/h3&gt;\u0000 \u0000 &lt;p&gt;Prognostic models can benefit from integrating heterogeneous data sources. While deep learning-based multimodal models often improve predictive performance, their limited interpretability remains a barrier to clinical adoption. Moreover, achieving robust generalization with such models typically requires large and diverse datasets, which are not always available in clinical research. This study evaluated whether a late fusion model combining traditional statistical approaches with deep learning models could rival deep learning-only multimodal models while preserving interpretability to support clinical use.&lt;/p&gt;\u0000 &lt;/section&gt;\u0000 \u0000 &lt;section&gt;\u0000 \u0000 &lt;h3&gt; Methods&lt;/h3&gt;\u0000 \u0000 &lt;p&gt;We conducted a retrospective single-center cohort study of 965 hospitalized patients with laboratory-confirmed SARS-CoV-2 infection who underwent frontal chest X-ray imaging during hospitalization between March 2020 and January 2022. Of these, 64 (6.6%) died. Structured clinical data and chest X-ray images were used to predict COVID-19-related death. We developed unimodal baseline models and two multimodal approaches: a deep learning-based joint fusion model and a late fusion model leveraging logistic regression to combine predictions from the best-performing models for every data modality. Given the limited sample size and class imbalance, performance was evaluated using repeated five-fold cross validation.&lt;/p&gt;\u0000 &lt;/section&gt;\u0000 \u0000 &lt;section&gt;\u0000 \u0000 &lt;h3&gt; Results&lt;/h3&gt;\u0000 \u0000 &lt;p&gt;The late fusion model achieved the best overall performance, with the highest median test AUPRC of 0.363 (IQR 0.271–0.384) across cross validation folds, outperforming unimodal baselines based on structured clinical data or images alone. The late fusion model also outperformed the joint fusion model, achieving, respectively, a 3.1% and 3.6% relative improvement in AUPRC and Brier score on pooled out-of-sample predictions and yielding the highest balanced F1-score. Moreover, it preserved partial interpretability by linking predictions to clinically meaningful variables such as age and comorbidities and enabling image-based predictions to be traced back to radiographic abnormalities.&lt;/p&gt;\u0000 &lt;/section&gt;\u0000 \u0000 &lt;section&gt;\u0000 \u0000 &lt;h3&gt; Conclusions&lt;/h3&gt;\u0000 \u0000 &lt;p&gt;Late fusion models integrating statistical models for clinical data with deep learning for image classification can enhance prognostic performance over unimodal models without fully sacrificing interpretability. In settings with limited data, they may also outperform deep learning-based joint fusion models. These findings support late fusion as a promising strategy that warrants prospecti","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"43 11","pages":""},"PeriodicalIF":3.4,"publicationDate":"2026-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1111/exsy.70440","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149048481","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":"ArGemma: A Multi-Task Fine-Tuning Framework for Adapting Gemma to Arabic","authors":"Taha Alselwi,&nbsp;Serap Çakar Kaman,&nbsp;Beyza Eken","doi":"10.1111/exsy.70431","DOIUrl":"https://doi.org/10.1111/exsy.70431","url":null,"abstract":"<p>Open-source large language models (LLMs) have significantly advanced natural language processing (NLP), particularly for English. However, their performance in Arabic has remained limited due to the scarcity of high-quality datasets and the high computational cost of full fine-tuning. To address this challenge, we curated and prepared a diverse collection of Arabic-specific datasets covering translation, summarization, storytelling, dialogue generation, and Classical–Modern Arabic bridging. Building on these resources, we developed ArGemma, an Arabic-adapted version of the Gemma model, fine-tuned using Supervised Fine-Tuning (SFT) with Low-Rank Adaptation (LoRA) to enable efficient adaptation with reduced computational overhead. A hybrid approach integrating SFT, Retrieval-Augmented Generation (RAG), and few-shot prompting was also employed to enhance contextual understanding and performance—particularly for tasks requiring nuanced reasoning across different forms of Arabic. ArGemma's performance was evaluated using both quantitative metrics (BLEU, ROUGE) and qualitative assessments by native speakers and language experts to measure fluency, coherence, and contextual accuracy. Results demonstrate that LoRA-based fine-tuning, supported by carefully prepared datasets and retrieval-augmented techniques, can significantly enhance Arabic NLP performance while remaining computationally efficient. This work, which was awarded third place in the 2025 Google—Unlock Global Communication with Gemma Kaggle competition, demonstrates how parameter-efficient fine-tuning (PEFT) combined with targeted dataset preparation can effectively expand LLM support for underrepresented languages like Arabic.</p>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"43 11","pages":""},"PeriodicalIF":3.4,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1111/exsy.70431","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149048153","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":"Heterogeneity-Aware Federated Learning via Heterogeneity Identification and Adaptive Optimization","authors":"Juncheng Pu,&nbsp;Xiaodong Fu,&nbsp;Li Liu,&nbsp;Wei Peng,&nbsp;Jiaman Ding","doi":"10.1111/exsy.70428","DOIUrl":"https://doi.org/10.1111/exsy.70428","url":null,"abstract":"<div>\u0000 \u0000 <p>Data heterogeneity remains a major challenge in building robust and scalable federated learning systems for real-world distributed applications, while many existing methods are designed for predefined heterogeneity settings and therefore have limited adaptability when multiple types of heterogeneity coexist across participants. To enable heterogeneity-aware federated learning, we propose an integrated framework that combines parameter-oriented heterogeneity identification with adaptive optimization. Specifically, a Transformer-based classifier identifies the heterogeneity type of each participant using uploaded model parameters. The attention weights learned during identification are further interpreted as heterogeneity-related parameter sensitivities and reused to guide subsequent optimization. For local updating, we introduce an attention-guided prior loss that imposes stronger constraints on heterogeneity-sensitive parameter regions while preserving local adaptability in less sensitive regions. For global aggregation, a weighted Gaussian-product scheme is employed to approximate the global posterior and mitigate aggregation errors caused by mixed heterogeneous data. The proposed identification method achieves a test accuracy of 98.22%. Extensive experiments on benchmark datasets and real-world medical image classification tasks demonstrate the effectiveness of the proposed framework under mixed data heterogeneity. Furthermore, sparse parameter exchange based on the attention weights reduces communication cost compared with FedAVG, supporting the practical applicability of the proposed framework in heterogeneous distributed learning environments.</p>\u0000 </div>","PeriodicalId":51053,"journal":{"name":"Expert Systems","volume":"43 11","pages":""},"PeriodicalIF":3.4,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149047913","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}