Information Fusion最新文献

{"title":"Poisoning attacks resilient privacy-preserving federated learning scheme based on lightweight homomorphic encryption","authors":"Chong Zhang ,&nbsp;Xiaojun Zhang ,&nbsp;Xingchun Yang ,&nbsp;Bingyun Liu ,&nbsp;Yuan Zhang ,&nbsp;Rang Zhou","doi":"10.1016/j.inffus.2025.103131","DOIUrl":"10.1016/j.inffus.2025.103131","url":null,"abstract":"<div><div>Federated learning ensures that multiple participants train the same model without leaking the local raw data. Each participant uploads the local gradient model instead of the original data, however, the uploaded local gradient model may contain certain sensitive information, which can be exploited by an adversary to break privacy protection. Meanwhile, some adversaries can make the model training results contrary to the expected results by tampering with the uploaded local gradient model or mixing malicious data into the local dataset, thereby inducing the model to produce wrong results for specific data. To this end, we devise a privacy-preserving federated learning scheme based on lightweight homomorphic encryption, which simultaneously reduces the weight of malicious data in gradient aggregation and supports anomaly detection of data, achieves the effect of resistance to poisoning attacks. Through theoretical analysis and experimental simulation, the proposed scheme has lightweight computation advantages compared with existing federated learning schemes.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103131"},"PeriodicalIF":14.7,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143740001","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":"PGA-DRL: Progressive graph attention-based deep reinforcement learning for recommender systems","authors":"Jawad Tanveer ,&nbsp;Sang-Woong Lee ,&nbsp;Amir Masoud Rahmani ,&nbsp;Khursheed Aurangzeb ,&nbsp;Mahfooz Alam ,&nbsp;Gholamreza Zare ,&nbsp;Pegah Malekpour Alamdari ,&nbsp;Mehdi Hosseinzadeh","doi":"10.1016/j.inffus.2025.103167","DOIUrl":"10.1016/j.inffus.2025.103167","url":null,"abstract":"<div><div>Advanced graph models, including Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs), have demonstrated their effectiveness in capturing intricate user-item interactions. However, their integration into Deep Reinforcement Learning (DRL)-based Recommender Systems (RSs) remains relatively underexplored. To address this gap, we propose PGA-DRL, a Progressive Graph Attention-Based DRL model that incrementally fuses GCN and GAT representations via concatenation, effectively combining their complementary strengths to enhance feature representation within an Actor-Critic (AC) framework. This progressive integration refines both global and localized user-item interaction patterns, Specifically, global patterns capture broader user preferences across the entire graph, and localized patterns focus on specific, detailed interactions between closely connected nodes, enabling a more comprehensive understanding of the recommendation environment. We evaluate our approach using extensive experiments on multiple benchmark datasets, including ML-100K, ML-1M, Amazon Subscription Boxes, Amazon Magazine Subscriptions, and ModCloth, employing standard ranking metrics such as Precision@10, Recall@10, NDCG@10, MRR@10, and Hit@10. The experimental results reveal that PGA-DRL outperforms state-of-the-art baselines, such as BPR, NeuMF, and SimGCL, achieving improvements in NDCG@10 and Recall@10. Our core contributions lie in bridging graph-based learning with reinforcement learning through a novel, efficient, and scalable fusion mechanism that enhances recommendation accuracy and ultimately improves user satisfaction. The source code for PGA-DRL is publicly available at <span><span>https://github.com/RS-Research/PGA-DRL</span><svg><path></path></svg></span> to enhance transparency and facilitate future research.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103167"},"PeriodicalIF":14.7,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143759323","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":"Divide and conquer? A combination of judgments method for comparing DSSs. Pairwise comparison vs. holistic paradigms","authors":"Carlos Sáenz-Royo ,&nbsp;Francisco Chiclana","doi":"10.1016/j.inffus.2025.103157","DOIUrl":"10.1016/j.inffus.2025.103157","url":null,"abstract":"<div><div>Despite the prevalence of Decision Support Systems (DSSs) in the field of decision-making, there is a paucity of research dedicated to the evaluation and comparison of these systems. This paper put forward a novel approach to symbolically encoding a DSS, which enables the generalization of comparisons between DSSs for any distribution of performances of the alternatives. The only hypothesis required in the proposed methodology is that the probability of choosing each alternative is proportional to its latent performance. The approach developed is demonstrated with its application to compare two paradigms commonly employed in DSS: holistic versus pairwise. Using a set of three alternatives, the present study provides mathematical proof that a DSS based on the pairwise comparison paradigm achieves higher expected performance than a DSS based on the holistic evaluation paradigm. This result challenges the emerging preference for holistic evaluation of alternatives and suggests that this result may apply to any number of alternatives.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103157"},"PeriodicalIF":14.7,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143768935","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":"Exploring EEG and eye movement fusion for multi-class target RSVP-BCI","authors":"Xujin Li ,&nbsp;Wei Wei ,&nbsp;Kun Zhao ,&nbsp;Jiayu Mao ,&nbsp;Yizhuo Lu ,&nbsp;Shuang Qiu ,&nbsp;Huiguang He","doi":"10.1016/j.inffus.2025.103135","DOIUrl":"10.1016/j.inffus.2025.103135","url":null,"abstract":"<div><div>Rapid Serial Visual Presentation (RSVP)-based Brain-Computer Interfaces (BCIs) enable high-throughput target image detection by identifying event-related potentials (ERPs) in electroencephalography (EEG) signals. Traditional RSVP-BCI systems detect only single-class targets within image streams, limiting their ability to handle more complex tasks requiring multi-class target identification. Multi-class target RSVP-BCI systems are designed to detect multi-class targets in real-world scenarios. However, distinguishing between different target categories remains challenging due to the high similarity across ERPs evoked by different target categories. In this work, we incorporate the eye movement (EM) modality into traditional EEG-based RSVP decoding and develop an open-source multi-modal dataset comprising EM and EEG signals from 43 subjects in three multi-class target RSVP tasks. We further propose the <strong>M</strong>ulti-class <strong>T</strong>arget <strong>R</strong>SVP <strong>E</strong>EG and <strong>E</strong>M fusion <strong>Net</strong>work (MTREE-Net) to enhance multi-class RSVP decoding. Specifically, a dual-complementary module is designed to strengthen the differentiation of uni-modal features across categories. To achieve more effective multi-modal fusion, we adopt a dynamic reweighting fusion strategy guided by theoretically derived modality contribution ratios for optimization. Furthermore, we propose a hierarchical self-distillation module to reduce the misclassification of non-target samples through knowledge transfer between two hierarchical classifiers. Extensive experiments demonstrate that MTREE-Net achieves significant performance improvements, including over 5.4% and 3.32% increases in balanced accuracy compared to existing EEG decoding and EEG-EM fusion methods, respectively. Our research offers a promising framework that can simultaneously detect target existence and identify their specific categories, enabling more robust and efficient applications in scenarios such as multi-class target detection.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103135"},"PeriodicalIF":14.7,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143735106","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":"Multilevel feature encoder for transfer learning-based fault detection on acoustic signal","authors":"Dezheng Wang ,&nbsp;Congyan Chen","doi":"10.1016/j.inffus.2025.103128","DOIUrl":"10.1016/j.inffus.2025.103128","url":null,"abstract":"<div><div>The intelligent diagnosis of faults in industrial assets is crucial for preventing unexpected disruptions to critical services. Although numerous deep learning methods based on acoustic data have been developed to enhance fault detection accuracy, these methods often prove suboptimal in transfer learning due to two key challenges: (1) insufficient generalization capability that causes overfitting to source device characteristics, and (2) failure to capture domain-invariant patterns essential for cross-device fault detection. This work seeks to alleviate these limitations by proposing a multilevel features encoder (MLFE) for transfer learning-based fault detection on acoustic signal. The acoustic data are initially preprocessed with a frequency mask to filter out high-frequency noise. Subsequently, feature engineering techniques are employed to extract several statistical features, such as the mean, standard deviation, and median absolute deviation, etc. with an emphasis on frequency characteristics. Moreover, unsupervised method is then applied to extract additional essential features. These multilevel features are then combined and fed into MLFE to differentiate between faulty and non-faulty signals. After being trained on several source devices, the pre-trained MLFE is transferred to a new target device to evaluate its transfer learning capability. MLFE is evaluated using the pump and fan datasets in MIMII, where it outperforms existing methods and offers a novel solution for transfer learning-based fault detection using acoustic signals.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103128"},"PeriodicalIF":14.7,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739987","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":"STATIC-LIO: A sliding window and terrain-assisted dynamic points removal LiDAR Inertial Odometry","authors":"Xuzhe Duan ,&nbsp;Qingwu Hu ,&nbsp;Mingyao Ai ,&nbsp;Pengcheng Zhao ,&nbsp;Meng Wu ,&nbsp;Jiayuan Li ,&nbsp;Chao Xiong","doi":"10.1016/j.inffus.2025.103132","DOIUrl":"10.1016/j.inffus.2025.103132","url":null,"abstract":"<div><div>With the development of diverse Light Detection and Ranging (LiDAR) sensors, LiDAR-based localization and mapping has become an essential issue in the fields of robotics and autonomous driving. However, the moving object in dynamic environments often introduces errors in LiDAR localization and leaves undesirable traces in the point cloud map. In this work, we propose a novel LiDAR inertial odometry (LIO) framework named STATIC-LIO, Sliding window and Terrain AssisTed dynamIC points removal LiDAR Inertial Odometry, which fuses the geometric, terrestrial, and motion information to enhance the localization and mapping performance. The terrestrial information is extracted through a fast progressive ground segmentation module designed to be compatible with various LiDARs. With the assistance of the terrestrial information, an online dynamic point voting mechanism is proposed to determine the motion information and remove the dynamic points in a point-wise manner. The ground segmentation and dynamic points removal modules are coupled within the sliding window-based STATIC-LIO framework to estimate odometry by leveraging geometric correspondences from ground and static points. We extensively evaluate the proposed framework on both public and real-world datasets encompassing a variety of LiDAR types. The experimental results demonstrate the effectiveness of STATIC-LIO across various datasets and applications, showcasing its superior accuracy by reducing localization errors by up to 92.4% compared to the state-of-the-art LIO framework.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103132"},"PeriodicalIF":14.7,"publicationDate":"2025-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143715390","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":"AMF-VSN: Adaptive multi-process fusion video steganography based on invertible neural networks","authors":"Yangwen Zhang ,&nbsp;Yuling Chen ,&nbsp;Hui Dou ,&nbsp;Yan Meng ,&nbsp;Haojin Zhu","doi":"10.1016/j.inffus.2025.103130","DOIUrl":"10.1016/j.inffus.2025.103130","url":null,"abstract":"<div><div>The security challenges in information transmission have attracted considerable research focus, particularly in the field of video steganography. Although deep learning advancements have created new research opportunities for video steganography, current models encounter deployment difficulties on mobile devices due to their substantial parameter requirements, which restrict their adaptability to mobile platform constraints. To address these challenges, we propose an adaptive multi-process fusion video steganography based on invertible neural networks. This model incorporates flexible balance adjustment factor and Multi Cross Stage Partial Dense (MCSPDense) Block, which adjusts the parameter count of the MCSPDense Block and the quality of the stego video through the flexible balance adjustment factor. Additionally, the Simple Redundancy Prediction Module (SRPM) has been designed to further minimize model parameters while enhancing the quality of video steganography and restoration. Furthermore, we develop two operational modes to accommodate varying mobile device requirements: Secure Communication (SC) mode for enhanced transmission security and High Quality Recovery (HQR) mode for superior video restoration. Experimental results confirm that compared to existing solutions, our AMF-VSN framework has improved steganography and recovery performance by 3.474 dB and 2.521 dB respectively, reduced parameters by 66.08%, and maintained strong security in mobile deployment scenarios.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103130"},"PeriodicalIF":14.7,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739986","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":"Reflections and attentiveness on eXplainable Artificial Intelligence (XAI). The journey ahead from criticisms to human–AI collaboration","authors":"Francisco Herrera","doi":"10.1016/j.inffus.2025.103133","DOIUrl":"10.1016/j.inffus.2025.103133","url":null,"abstract":"<div><div>The emergence of deep learning over the past decade has driven the development of increasingly complex AI models, amplifying the need for Explainable Artificial Intelligence (XAI). As AI systems grow in size and complexity, ensuring interpretability and transparency becomes essential, especially in high-stakes applications. With the rapid expansion of XAI research, addressing emerging debates and criticisms requires a comprehensive examination. This paper explores the complexities of XAI from multiple perspectives, proposing six key axes that shed light on its role in human–AI interaction and collaboration. First, it examines the imperative of XAI under the dominance of black-box AI models. Given the lack of definitional cohesion, the paper argues that XAI must be framed through the lens of audience and understanding, highlighting its different uses in AI–human interaction. The recent BLUE vs. RED XAI distinction is analyzed through this perspective. The study then addresses the criticisms of XAI, evaluating its maturity, current trajectory, and limitations in handling complex problems. The discussion then shifts to explanations as a bridge between AI models and human understanding, emphasizing the importance of usability of explanations in human–AI decision making. Key aspects such as AI reliance, human intuition, and emerging collaboration theories — including the human-algorithm centaur and co-intelligence paradigms — are explored in connection with XAI. The medical field is considered as a case study, given its extensive research on collaboration between doctors and AI through explainability. The paper proposes a framework to evaluate the maturity of XAI using three dimensions: practicality, auditability, and AI governance. Provide the final lessons learned focused on trends and questions to tackle in the near future. This is an in-depth exploration of the impact and urgency of XAI in the era of pervasive expansion of AI. Three Key reflections from this study include: (a) XAI must enhance cognitive engagement with explanations, (b) it must evolve to fully address why, what, and for what purpose explanations are needed, and (c) it plays a crucial role in building societal trust in AI. By advancing XAI in these directions, we can ensure that AI remains transparent, auditable, and accountable, and aligned with human needs.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103133"},"PeriodicalIF":14.7,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143739985","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":"PPSTSL: A Privacy-preserving Dynamic Spatio-temporal Graph Data Federated Split Learning for traffic forecasting","authors":"Yan Feng ,&nbsp;Quan Qian","doi":"10.1016/j.inffus.2025.103129","DOIUrl":"10.1016/j.inffus.2025.103129","url":null,"abstract":"<div><div>In intelligent transportation, federated learning has garnered significant attention for its privacy protection and model optimization capabilities. However, existing approaches still struggle with high computational and communication costs. Moreover, challenges such as the uneven distribution of traffic nodes and time sequences, global dynamic spatio-temporal correlation, and security concerns in the spatio-temporal modeling process remain insufficiently addressed in distributed transportation research. To tackle these challenges, this study proposed a Privacy-Preserving Dynamic Spatio-temporal Graph Data Federated Split Learning (PPSTSL) method. By incorporating a split learning framework for spatio-temporal modeling under privacy protection, PPSTSL enables the integration of multi-client data to enhance global model performance while mitigating computational and communication overhead. Specifically, we present Federated Distribution-Aligned Temporal Dependency Modeling (FedDATDep) to optimize the model parameters and enrich temporal features. Additionally, we design Federated Spatio-Temporal Fusion (FedTSFus) to achieve global dynamic spatio-temporal dependencies while preserving privacy. Furthermore, we propose a Positive-Negative Coupled Coding (PNCC) mechanism to enhance the computational and communication efficiency of the introduced security techniques. Experimental results show that PPSTSL achieves superior model performance in both independent and non-independent identical distribution scenarios (IID and Non-IID) on the SZ-Taxi and Los-Loop traffic datasets, while demonstrating strong scalability. Furthermore, the optimized security techniques not only enhance privacy protection but also reduce communication and computational overhead. The proposed PPSTSL approach addresses crucial limitations in existing distributed research methods in intelligent transportation, presenting a promising direction for future research and practical implementations.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103129"},"PeriodicalIF":14.7,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143725257","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":"Triplets-based large-scale multi-view spectral clustering","authors":"Tianchuan Yang ,&nbsp;Chang-Dong Wang ,&nbsp;Jipeng Guo ,&nbsp;Xiangcheng Li ,&nbsp;Man-Sheng Chen ,&nbsp;Shuping Dang ,&nbsp;Haiqiang Chen","doi":"10.1016/j.inffus.2025.103134","DOIUrl":"10.1016/j.inffus.2025.103134","url":null,"abstract":"<div><div>By integrating complementary information from multiple views to reach consensus, graph-based multi-view clustering describes data structure competently, thereby attracting considerable attention. It is time-bottlenecked by graph construction and eigen-decomposition in the big data era. Existing methods usually utilize anchor graph learning to address this issue. However, problems of unsupervised representative anchor selection, consensus or view-specific anchors, anchor alignment, etc., remain challenging. Moreover, excessive information is discarded for the sake of efficiency. Motivated by the essence of the anchor-based methods that utilizing representative point-to-point relations to reduce graph complexity, we generate triplets for each view based on neighborhood similarity to preserve point-to-point relations and local structure, and propose triplets-based large-scale multi-view spectral clustering (TLMSC). Subsequently, the triplet enhancement strategy is designed to select representative triplet relations to improve efficiency and clustering performance. Specifically, positive examples of triplets are filtered according to the view consensus to significantly increase the probability of positive examples belonging to the same cluster. The most indistinguishable hard negative examples are generated based on probabilities to improve discrimination performance. Guided by the enhanced triplets and its weights, an improved low-dimensional embedding is constructed through optimization, which further serves as an input to the proposed fast sparse spectral clustering (FSSC) to obtain clustering results. Numerous experiments validate the efficiency and superior performance of the proposed TLMSC. An average improvement of 12.25% at least in ACC compared to 10 state-of-the-art methods on 18 datasets. The code is available at <span><span>github.com/ytccyw/TLMSC</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"121 ","pages":"Article 103134"},"PeriodicalIF":14.7,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143715392","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}