Information Fusion最新文献

{"title":"PrivLSTM: A privacy-preserving LSTM inference framework by fusing encryption and network structure for multi-sourced Data","authors":"Zhenyong Zhang ,&nbsp;Kuan Shao ,&nbsp;Ruilong Deng ,&nbsp;Xin Wang ,&nbsp;Yishu Zhang ,&nbsp;Mufeng Wang","doi":"10.1016/j.inffus.2025.103711","DOIUrl":"10.1016/j.inffus.2025.103711","url":null,"abstract":"<div><div>Machine Learning as a Service (MLaaS) has emerged as a prominent topic for dealing with multi-sourced data. However, the privacy concerns have received more considerable attention than ever. In this paper, we propose PrivLSTM, a low latency privacy-preserving long short-term memory (LSTM) neural network inference method by fusing the encrypted data and network structure, protecting the sequence data input to the LSTM. PrivLSTM is implemented with a non-interactive framework using fully homomorphic encryption and is adaptive to a general LSTM structure with long input sequences. PrivLSTM uses bootstrapping to avoid the impact of noise introduced by the deep multiplications on the accuracy. The computation overhead of bootstrapping is alleviated by developing a novel batch-based linear transformation method, which reduces the automorphism operations and integrates identical operations in the LSTM cell. The nonlinear operations, e.g., sigmoid and tanh, are approximated with optimized polynomials. Besides, the efficiency and security of PrivLSTM are theoretically analyzed. We conduct experiments on multi-sourced datasets, such as text and sensory datasets. The results show that PrivLSTM achieves comparable accuracy against the plain LSTM and fast computation with around 30s for a 100-length input sequence.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103711"},"PeriodicalIF":15.5,"publicationDate":"2025-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107898","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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 kernelized fuzzy approximation fusion model with granular-ball computing for outlier detection","authors":"Yongxiang Li ,&nbsp;Xinyu Su ,&nbsp;Zhong Yuan ,&nbsp;Run Ye ,&nbsp;Dezhong Peng ,&nbsp;Hongmei Chen","doi":"10.1016/j.inffus.2025.103716","DOIUrl":"10.1016/j.inffus.2025.103716","url":null,"abstract":"<div><div>Outlier detection is a fundamental task in data analytics, where fuzzy rough set-based methods have gained increasing attention for their ability to effectively model uncertainty associated with outliers in data. However, existing FRS-based methods often exhibit limitations when applied to complex scenarios. Most of these methods rely on single-granularity fusion, where all samples are processed at a uniform, fine-grained level. This restricts their ability to fuse multi-granularity information, limiting outlier discrimination and making them more susceptible to noise. Moreover, many traditional methods construct fuzzy relation matrices under linear assumptions, which fail to effectively represent the intricate, nonlinear relations commonly found in real-world data. This leads to suboptimal estimation of membership degrees and degrades the reliability of outlier detection. To address these challenges, we propose a Kernelized Fuzzy approximation fusion model with Granular-ball computing for Outlier Detection (KFGOD), which integrates multi-granularity granular-balls and kernelized fuzzy rough sets into a unified framework. KFGOD fuses multi-granularity information to capture abnormal information at different granularity levels. Simultaneously, kernel functions are employed to effectively model multi-granularity nonlinear relations, enhancing the expressive power of fuzzy relations. By performing information fusion across multiple kernelized fuzzy information granules associated with each granular-ball, KFGOD evaluates the outlier degrees of each ball and propagates this fused abnormality information to the corresponding samples. This hierarchical and kernelized method allows for effective outlier detection in unlabeled datasets. Extensive experiments conducted on twenty benchmark datasets confirm the effectiveness of KFGOD, which consistently outperforms several state-of-the-art baselines in terms of detection accuracy and robustness. The codes are publicly available online at <span><span>https://github.com/LYXRhythm/KFGOD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103716"},"PeriodicalIF":15.5,"publicationDate":"2025-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107555","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Multi-granularity interval-intent fuzzy concept-cognitive learning: An attention-enhanced adaptive clustering framework","authors":"Yi Ding,&nbsp;Weihua Xu","doi":"10.1016/j.inffus.2025.103715","DOIUrl":"10.1016/j.inffus.2025.103715","url":null,"abstract":"<div><div>Cognitive processes lie at the heart of artificial intelligence (AI) research, and the Multi-Granularity Interval-Intent Fuzzy Concept-Cognitive Learning model (MIFCL-A) presented in this paper offers a novel perspective on this domain. MIFCL-A innovatively incorporates multi-level attention mechanism to replicate the intricacies of human cognition, utilizing advanced concept cognitive learning methodologies. This model addresses several limitations inherent in existing concept learning frameworks, such as reliance on manual parameter tuning for concept clustering, the generation of pseudo concepts that compromise cognitive consistency, and an overreliance on attribute-based concept attention that neglects the centrality of objects. Our model introduces a multi-granularity concept structure that captures both global (coarse-granularity) and local (fine-granularity) perspectives, integrating global decision concepts with boundary-derived local concepts. It features a hierarchical attention mechanism that applies global attribute attention at the coarse-granularity level and local concept attention at the fine-granularity level. Moreover, an adaptive concept clustering algorithm is incorporated, which negates the need for manual parameter tuning and ensures the precision and robustness of concept evolution across varying granularities. Comparative evaluations indicate that MIFCL-A outperforms current models in terms of classification accuracy and knowledge representation capabilities, establishing its potential as an effective tool for knowledge discovery and data mining.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103715"},"PeriodicalIF":15.5,"publicationDate":"2025-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107896","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"ChronoSentinel: Incremental temporal embedding for Security Knowledge Graph using Dynamic Reachability Centrality and Efficient language model","authors":"Chinmaya Mishra ,&nbsp;Himangshu Sarma ,&nbsp;Saravanan M.","doi":"10.1016/j.inffus.2025.103662","DOIUrl":"10.1016/j.inffus.2025.103662","url":null,"abstract":"<div><div>The increasing sophistication of cyber threats requires adaptive, real-time defenses that can evolve with dynamic attack patterns. Security Knowledge Graphs (SKGs) have become essential for representing complex interrelationships among cyber entities, which are vital for combating ongoing cybercrime. However, most existing incremental update methods rely on non-temporal strategies that fail to capture the evolution of security data. This paper presents ChronoSentinel, an innovative framework that synergistically integrates Dynamic Reachability Centrality (DRC) with Efficient Language Models (ELMs) to offer a robust and scalable solution for maintaining and enhancing Temporal Security Knowledge Graphs. By incorporating temporal dynamics, ChronoSentinel incrementally updates the graph while reducing the computational cost of full retraining, leveraging time-sensitive information to respond to emerging threats. The framework employs DRC to prioritize influential and temporally critical core nodes, ensuring the graph remains up-to-date and responsive to evolving threat landscapes. Additionally, by integrating ELMs such as BART, FLAN-T5, and DeepSeek, ChronoSentinel enriches the graph with contextual insights that improve semantic representation and enable predictive link generation. This hybrid approach supports faster threat prediction and defense while maintaining reliability, accuracy, and low computational overhead.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103662"},"PeriodicalIF":15.5,"publicationDate":"2025-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145057398","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Aggregation-based self-supervised evidential clustering for imbalanced data","authors":"Zuowei Zhang ,&nbsp;Hongpeng Tian ,&nbsp;Jingwei Zuo ,&nbsp;Weiping Ding","doi":"10.1016/j.inffus.2025.103721","DOIUrl":"10.1016/j.inffus.2025.103721","url":null,"abstract":"<div><div>Clustering, as a fusion process, involves aggregating similar objects and isolating dissimilar ones, independent of any prior information. Recently, evidential clustering has gained popularity due to its ability to characterize the uncertainty and imprecision of data distribution. However, it remains a major bottleneck of existing evidential clustering methods for clustering imbalanced data, as they cannot effectively detect small clusters (with a few objects). In this paper, we propose a new aggregation-based self-supervised evidential clustering (ASEC) method for dealing with such issues based on the theory of belief functions. Specifically, a cluster density-based aggregation rule is designed first to generate multiple sub-clusters and then fuse them into new singleton clusters, which can effectively detect small clusters of imbalanced data. The new singleton clusters obtained by the aggregation rule serve as prior knowledge. Then, a self-supervised evidential partition rule is developed to fuse the remaining objects into new clusters according to prior knowledge and the <span><math><mi>K</mi></math></span>-nearest neighbors (KNNs) technique. In this process, the objects in the overlapping zones of clusters are usually hard to classify, and they are assigned to new meta-clusters to reduce the risk of error. Experiments on several imbalanced datasets demonstrate the effectiveness of ASEC compared to related methods.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103721"},"PeriodicalIF":15.5,"publicationDate":"2025-09-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107895","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Fusion techniques of frame and event cameras in autonomous driving: A review","authors":"Peijun Shi,&nbsp;Chee-Onn Chow,&nbsp;Wei Ru Wong","doi":"10.1016/j.inffus.2025.103697","DOIUrl":"10.1016/j.inffus.2025.103697","url":null,"abstract":"<div><div>The rapid advancement of autonomous driving technology demands robust environmental perception systems capable of operating under extreme illumination variations, high-speed motion, and adverse weather conditions. While conventional frame-based cameras offer rich spatial and textural information, they suffer from fixed frame rates and limited dynamic range. Event cameras, as neuromorphic vision sensors, provide unique advantages in temporal resolution, dynamic range, and power efficiency through asynchronous pixel-level brightness change detection. This paper presents the first comprehensive review of frame-event camera fusion technology for autonomous driving applications. This paper establishes a fusion framework tailored to autonomous driving perception requirements and analyzes the complementary characteristics of both sensor modalities in addressing critical perception challenges. This paper proposes the first systematic classification of frame-event fusion architectures, covering multi-level approaches from data-level to decision-level integration, while tracing the technical evolution of fusion strategies. Additionally, this paper constructs a dataset evaluation framework for autonomous driving tasks, providing systematic benchmark selection guidance. Through detailed analysis of deployment challenges, this paper identifies key technical barriers including temporal synchronization, computational efficiency, and cross-modal calibration, alongside corresponding solutions. Finally, this paper presents perspectives on emerging paradigms and future directions, providing essential references for advancing practical frame-event fusion applications in autonomous driving.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103697"},"PeriodicalIF":15.5,"publicationDate":"2025-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107553","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"DSFNet: Dual-fusion network with secondary clustering and feature integration for unsupervised person re-identification","authors":"Mingfu Xiong ,&nbsp;Abdul Khader Jilani Saudagar ,&nbsp;Mohammad Hijji ,&nbsp;Khan Muhammad ,&nbsp;Muhammad Haris Khan","doi":"10.1016/j.inffus.2025.103701","DOIUrl":"10.1016/j.inffus.2025.103701","url":null,"abstract":"<div><div>Unsupervised person re-identification (ReID) aims to train a model by updating a memory dictionary to retrieve a person of interest from different camera views. This area has attracted widespread interest recently due to its low cost of data processing. Existing methods predominantly concentrate on utilizing a single, fixed optimization approach (network) for clustering all features, neglecting their diversity and integrity, thereby resulting in noisy pseudo-labels. To solve this problem, this study proposes a DSFNet framework, namely, a Dual-fusion network with secondary clustering and feature integration (DSFNet) framework for unsupervised person ReID tasks. The proposed framework consists of three main components: (i) a hard-sample secondary clustering network (SCNet), (ii) a feature integration network (FINet), and (iii) a Dual-fusion dynamic optimization (DDO) scheme. Specifically, the first module focuses on initializing the memory dictionary via a hard-sample (dissimilar samples in the intra-class or similar samples in the inter-class) secondary clustering strategy, which preserves the diversity of the individual features. The FINet explores each person’s local and global features via an integrated weight-assigned strategy to ensure the integrity of the individual instance features. Next, the dual-fusion dynamic optimization scheme is implemented from FINet to SCNet, thereby guaranteeing consistent clustering features and ultimately mitigates the noise related to the generation of pseudo-labels. Extensive experimental results demonstrate that, compared to the latest pure unsupervised methods on commonly used ReID datasets (such as Market-1501, MSMT17, and PersonX), our approach achieves performance improvements in mAP of 1.7%, 1.9% and 4.0%, respectively. Similarly, Rank@1 performance has seen improvements of 0.1%, 4.6% and 1.0%, respectively. Meanwhile, to verify the generalization capability of our method, we conducted tests on the additional vehicle re-identification dataset VeRi-776, which exhibited performance largely comparable to the latest methods.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103701"},"PeriodicalIF":15.5,"publicationDate":"2025-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145107556","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Modality divergence based evolutionary learning for unsupervised visible-infrared cross-modality person re-identification","authors":"Yuxuan Liu ,&nbsp;Hongwei Ge ,&nbsp;Yong Luo ,&nbsp;Chunguo Wu","doi":"10.1016/j.inffus.2025.103706","DOIUrl":"10.1016/j.inffus.2025.103706","url":null,"abstract":"<div><div>Unsupervised visible-infrared cross-modality person re-identification aims to learn cross-modality invariant features between visible and infrared modalities without relying on labeled data. Currently, the state-of-the-art methods optimize cross-modality differences by reducing intra-class gaps while expanding inter-class gaps as the underlying paradigm. However, since the cross-modality intra-class gaps are huge, there must be a large number of inter-class instances between the gaps, and such inter-class instances make cross-modality intra-class instances difficult to get closer to each other in the feature space. To this end, we propose a modality divergence based evolutionary learning framework to optimize the cross-modality intra- and inter-class instance distribution. Specifically, on the one hand, we explore the optimization directions of each cluster in two modalities and make the explored attack and defense clusters perform mutual adversarial evolutionary learning through selection, crossover, and mutation, which produces the optimal inter-class distribution. On the other hand, we explore the intra-class instances with maximum and minimum similarity and perform mutual evolutionary optimization between the maximum and minimum instances, which retains only the modality changes in the intra-class instances to learn cross-modality invariant features. Extensive experiments conducted on datasets for visible-infrared person re-identification demonstrate that the proposed approach outperforms current state-of-the-art methods.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"126 ","pages":"Article 103706"},"PeriodicalIF":15.5,"publicationDate":"2025-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145048607","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"HandDiff: Spatial–temporal diffusion model for hand pose forecasting","authors":"Jinguang Tong ,&nbsp;Kaihao Zhang","doi":"10.1016/j.inffus.2025.103647","DOIUrl":"10.1016/j.inffus.2025.103647","url":null,"abstract":"<div><div>We propose a novel problem of forecasting future 3D hand pose from a short past sequence. The primary challenge in this task is accurately modeling the stochastic nature of future hand movements. To address this, we propose a diffusion-based hand pose forecasting model designed to generate accurate future hand poses by leveraging spatial–temporal information. Our model incorporates a Spatial–Temporal Attention Module (STAM) to capture correlations between hand joints and time points, and a Coarse Forecasting Module (CFM) to extract limited explicit guidance from the temporal dimension. These features condition the diffusion model to forecast plausible future hand poses. Due to the lack of suitable datasets, we also construct two large-scale datasets based on the existing hand-object interaction (HOI) datasets HO-3D and HOI4D for benchmarking hand pose forecasting, covering both third-person and egocentric perspectives. Experimental results show that our method HandDiff significantly outperforms other state-of-the-art (SOTA) methods by 16.7% on the HO-3D dataset and 11.1% on the HOI4D dataset in terms of the mean per joint position error (MPJPE), respectively.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"127 ","pages":"Article 103647"},"PeriodicalIF":15.5,"publicationDate":"2025-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145061293","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"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":"Spatio-temporal data fusion framework based on large language model for enhanced prediction of electric vehicle charging demand in smart grid management","authors":"Yitong Shang ,&nbsp;Wen-Long Shang ,&nbsp;Dingsong Cui ,&nbsp;Peng Liu ,&nbsp;Haibo Chen ,&nbsp;Dongdong Zhang ,&nbsp;Runsen Zhang ,&nbsp;Chengcheng Xu ,&nbsp;Ye Liu ,&nbsp;Chenxi Wang ,&nbsp;Mohannad Alhazmi","doi":"10.1016/j.inffus.2025.103692","DOIUrl":"10.1016/j.inffus.2025.103692","url":null,"abstract":"<div><div>Accurate prediction of electric vehicle (EV) charging demand is pivotal for effective smart grid management and renewable energy integration. However, predicting spatio-temporal EV charging patterns remains challenging due to complex data fusion requirements arising from heterogeneous temporal, spatial, and contextual features, as well as difficulties in effectively integrating multiple modeling approaches. This paper introduces EV-STLLM, a novel spatio-temporal data fusion framework based on Large Language Model explicitly designed for accurate short-term EV charging demand forecasting through innovative integration of data-level and model-level fusion techniques. At the data level, a multi-source embedding module is developed to seamlessly fuse temporal features (e.g., time slots, weekdays), spatial heterogeneity (e.g., geographical location), and contextual charging behaviors into a unified representation via embedding convolutional network. At the model level, a large language model (LLM) is employed to capture global spatiotemporal dependencies, enhanced with Low-Rank Adaptation (LoRA) for parameter-efficient fine-tuning, substantially reducing computational costs while maintaining prediction robustness. Using a comprehensive real-world dataset comprising over 830,000 EV charging records across 16 districts and 331 subdistricts in Beijing, we validate EV-STLLM across multiple forecasting scenarios (district and subdistrict levels, one-step and two-step ahead predictions). Extensive comparative evaluations demonstrate that EV-STLLM consistently outperforms classical, graph-based, and deep learning baselines. Specifically, in one-step ahead district-level forecasting, EV-STLLM achieves up to a 15.41% reduction in MAE and a 53.51% reduction in MAPE compared to the leading baseline, underscoring its potential to significantly enhance data-driven smart grid operations.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"126 ","pages":"Article 103692"},"PeriodicalIF":15.5,"publicationDate":"2025-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145048606","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}