Information Fusion最新文献

TLKN: Transformer-Style Large Kernel Networks for Effective Feature Fusion and Efficient Classification of Hyperspectral Images TLKN：变压器式大核网络用于高光谱图像的有效特征融合和高效分类

IF 18.6 1区计算机科学

Information Fusion Pub Date : 2026-11-04 DOI: 10.1016/j.inffus.2026.104917

Xun Liu, Jinchang Ren, RongJun Chen, Alex Hay-Man Ng, Fangyuan Lei, Amir Hussain, Tariq Salim Durrani

{"title":"TLKN: Transformer-Style Large Kernel Networks for Effective Feature Fusion and Efficient Classification of Hyperspectral Images","authors":"Xun Liu, Jinchang Ren, RongJun Chen, Alex Hay-Man Ng, Fangyuan Lei, Amir Hussain, Tariq Salim Durrani","doi":"10.1016/j.inffus.2026.104917","DOIUrl":"https://doi.org/10.1016/j.inffus.2026.104917","url":null,"abstract":"Hybrid models combining Transformers with convolutions have been utilized for hyperspectral image classification (HSIC) to compensate for the limited local feature extraction ability of Transformer only architectures. However, both Transformer only and hybrid models are inherently constrained by the quadratic computational complexity due to the multi-head self-attention (MHSA) mechanisms adopted within the Transformer. To mitigate this limitation, we propose a novel Transformer-style large kernel network (TLKN) that achieves linear computational complexity while preserving the ability to capture short- and within-patch long-range spatial dependencies. Specifically, TLKN performs effective feature fusion through two complementary modules: a 3 × 3 convolutional module for capturing local spectral-spatial information, and a Transformer-style large kernel convolutional (TSLKC) module for extracting both short- and long-range dependencies. The key innovation lies in the large kernel convolution attention (LKCA) mechanism embedded within TSLKC, which replaces the linear projections and matrix multiplications in MHSA with multiple parallel asymmetric depthwise convolutions (ADCs) and element-wise additions, thereby reducing the computational complexity from quadratic to linear without compromising long-range dependency modeling. Extensive experiments on four HSI benchmarks have demonstrated that TLKN achieves state-of-the-art classification performance with superior efficiency. For instance, on the HongHu dataset, it outperforms two Transformer-only based methods by 19.04% and 35.33% and three hybrid Transformer-based methods by 13.96% to 18.79% in overall accuracy (OA), respectively. Moreover, it is more than six times faster than the Transformer-only methods and nearly 2.7 times faster than hybrid ones on average.","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"11 1","pages":""},"PeriodicalIF":18.6,"publicationDate":"2026-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149359111","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}

引用次数: 0

End-to-End Privacy in Brain–Computer Interfaces: A Survey of Vulnerabilities, Privacy-Enhancing Technologies, and Governance 脑机接口的端到端隐私：漏洞、隐私增强技术和治理的调查

IF 18.6 1区计算机科学

Information Fusion Pub Date : 2026-11-04 DOI: 10.1016/j.inffus.2026.104925

George Ford, Jingyuan Sun, Mustafa Mustafa, Zhenhong Li

{"title":"End-to-End Privacy in Brain–Computer Interfaces: A Survey of Vulnerabilities, Privacy-Enhancing Technologies, and Governance","authors":"George Ford, Jingyuan Sun, Mustafa Mustafa, Zhenhong Li","doi":"10.1016/j.inffus.2026.104925","DOIUrl":"https://doi.org/10.1016/j.inffus.2026.104925","url":null,"abstract":"Non-invasive brain–computer interfaces (BCIs) are extending from clinically supervised research systems into consumer platforms connected to smartphones, remote services, and third-party applications. This shift broadens the trust boundaries across which neural signals flow and enables secondary inference and biometric linkage beyond the task for which recording was authorised. This survey reframes BCI privacy as an end-to-end property of a distributed, closed-loop pipeline in which EEG may be combined with gaze, motion, and contextual telemetry. We make three contributions. First, we systematise the privacy threat landscape by mapping twelve attack classes to a generalised four-stage BCI architecture and its data lifecycle. We extend this analysis to observation-, feature-, and decision-level fusion, identifying how joint representations, alignment metadata, and recipients’ access change existing threats. Second, we develop the <ce:italic>BCI Privacy Trilemma</ce:italic>: the coupled requirements of privacy protection, decoding utility, and real-time responsiveness. Selected empirical reference configurations ground the comparison of privacy-enhancing technologies while making differences in tasks, hardware, and protection assumptions explicit. We also examine safeguards for the sensory-feedback channel and distinguish demonstrated components from proposed security combinations. Third, we propose the fourteen-item <ce:italic>BCI Privacy Reporting Standard (BPRS)</ce:italic>, with a concrete baseline evaluation proposal and guidance for unavailable measurements in proprietary systems. Together, the architecture, trilemma, and reporting standard support interpretable, comparable, and auditable assessment of consumer BCI privacy without treating checklist completion or any single mechanism as a whole-system guarantee.","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"15 1","pages":""},"PeriodicalIF":18.6,"publicationDate":"2026-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149359110","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}

引用次数: 0

Adaptive Dual-View Score Fusion for Open-Set Node Classification on Heterophilic Graphs 异亲图开集节点分类的自适应双视图评分融合

IF 18.6 1区计算机科学

Information Fusion Pub Date : 2026-11-03 DOI: 10.1016/j.inffus.2026.104921

Long Xu, Zhiqiang Pan, Honghui Chen

{"title":"Adaptive Dual-View Score Fusion for Open-Set Node Classification on Heterophilic Graphs","authors":"Long Xu, Zhiqiang Pan, Honghui Chen","doi":"10.1016/j.inffus.2026.104921","DOIUrl":"https://doi.org/10.1016/j.inffus.2026.104921","url":null,"abstract":"Open-set node classification methods uniformly rely on propagating out-of-distribution signals through graph topology, an assumption that fails on heterophilic graphs where connected nodes carry different labels. We formalize the out-of-distribution signal dilution effect: when cross-distribution edges dominate, neighborhood aggregation attenuates the discriminative gap between in-distribution and out-of-distribution scores. This produces three failure modes that share a single root cause, the blind aggregation of unreliable information sources. To address these failures, we propose HetOOD, a per-node adaptive propagation framework that decouples ego features from neighbor aggregation via learned gating, preventing representation collapse on heterophilic regions. Its dual-view out-of-distribution scoring module combines topology-free and topology-aware signals through a score-consensus weight derived from neighborhood scoring agreement, adapting propagation strength per node without labels. Experiments on 9 benchmarks show consistent improvements over all existing methods, particularly on the most challenging heterophilic datasets.","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"18 1","pages":""},"PeriodicalIF":18.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149359132","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}

引用次数: 0

DCRS: Attention-Guided Dual-Channel Reasoning Segmentation for Industrial Anomaly Detection DCRS：用于工业异常检测的注意力引导双通道推理分割

IF 18.6 1区计算机科学

Information Fusion Pub Date : 2026-11-03 DOI: 10.1016/j.inffus.2026.104927

Ruiyang Jin, Jingjie Zeng, Huayang Li, Zewen Bai, Hongfei Lin, Liang Yang

{"title":"DCRS: Attention-Guided Dual-Channel Reasoning Segmentation for Industrial Anomaly Detection","authors":"Ruiyang Jin, Jingjie Zeng, Huayang Li, Zewen Bai, Hongfei Lin, Liang Yang","doi":"10.1016/j.inffus.2026.104927","DOIUrl":"https://doi.org/10.1016/j.inffus.2026.104927","url":null,"abstract":"Industrial Anomaly Detection (IAD) serves as a core component in intelligent manufacturing, requiring accurate pixel-level localization of anomalies. Recently, Multimodal Large Language Models (MLLMs) have introduced semantic interaction capabilities into this field, enabling inspection systems to process natural language instructions. However, most vision-language integration solutions adopt a decoupled “perception–reasoning” paradigm, resulting in error propagation across stages, computational redundancy, and loss of spatial details required for boundary-aware localization. To address this, we propose DCRS: an attention-guided MLLM-based end-to-end reasoning segmentation framework. At its core, DCRS uses the generated <ce:monospace><SEG></ce:monospace> position as a shared cross-layer readout anchor and establishes a segmentation-intent-aligned explicit–implicit dual-channel interface. The explicit channel reads the final-layer <ce:monospace><SEG></ce:monospace> hidden state as a semantic target specification and maps it to SAM’s sparse prompt space, whereas the implicit channel mines transition-zone <ce:monospace><SEG></ce:monospace>-to-image attention and reshapes it into a continuous dense prompt. By coordinating these heterogeneous readouts before mask decoding, DCRS provides fine-grained spatial guidance without introducing a separate anomaly-map generation stage. Experiments on MVTec-AD and VisA demonstrate that DCRS achieves strong anomaly detection and pixel-level localization performance, together with effective synthetic-to-real transfer, consistent cross-dataset generalization, and stable behavior across backbone scales. Evaluation on MMAD further shows that DCRS retains effective industrial diagnostic understanding. Meanwhile, under the same inference setup, DCRS achieves a substantial observed end-to-end speedup over AnomalyGPT, demonstrating the favorable system-level inference efficiency of its unified reasoning-and-segmentation pipeline.","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"13 1","pages":""},"PeriodicalIF":18.6,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149359131","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}

引用次数: 0

Generative Intent Subgraph Augmentation for Context-Enhanced Recommendation 上下文增强推荐的生成意图子图增强

IF 18.6 1区计算机科学

Information Fusion Pub Date : 2026-11-02 DOI: 10.1016/j.inffus.2026.104926

Meng Jian, Guangtai Zhao, Chaoqun Niu, Lifang Wu, Qian Qiu

{"title":"Generative Intent Subgraph Augmentation for Context-Enhanced Recommendation","authors":"Meng Jian, Guangtai Zhao, Chaoqun Niu, Lifang Wu, Qian Qiu","doi":"10.1016/j.inffus.2026.104926","DOIUrl":"https://doi.org/10.1016/j.inffus.2026.104926","url":null,"abstract":"In the era of information overload, personalized recommendation systems have emerged as a pivotal technology for alleviating user burden in navigating vast information spaces. While the evolution from classical collaborative filtering to behavioral graph learning models has effectively mitigated the interaction sparsity problem, existing context-enhanced collaborative filtering models frequently suffer from redundancy and noise embedded in contextual descriptions, thereby hindering the capture of complex user interests. To address this challenge, this work leverages the superior semantic understanding and reasoning capabilities of large language models (LLMs) to depict users’ hidden intents in semantics and proposes a novel generative intent subgraph augmentation (GISA) model for recommendation. It aims to resolve contextual noise issues through semantic filtering and fusion to learn fine-grained interests. Specifically, GISA employs LLMs to generate explicit intent queries from the contextual texts of user interactions, which then prompt GISA to disentangle the complex interaction graph into multiple intent-specific subgraphs. Subsequently, GISA learns intent-aware embeddings by modeling interests on each intent subgraph and performs intent-aware semantic retrieval to augment the interaction graph, effectively alleviating the sparsity issue in collaborative filtering signals. Extensive experiments on four public datasets demonstrate that GISA consistently outperforms the representative baselines, validating the effectiveness of leveraging LLMs for understanding complex user intent compositions and guiding behavioral augmentation in recommendation systems.","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"22 1","pages":""},"PeriodicalIF":18.6,"publicationDate":"2026-11-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149359133","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}

引用次数: 0

ReT-UAP: Rethinking Multi-Surrogate Universal Adversarial Perturbations on Vision-Language Pre-trained Models ReT-UAP：对视觉语言预训练模型的多代理通用对抗性扰动的再思考

IF 18.6 1区计算机科学

Information Fusion Pub Date : 2026-11-02 DOI: 10.1016/j.inffus.2026.104913

Ziyu Wang, Jian Li, Yiming Du, Rui Ning, Jiang Li, Daniel Takabi, Lusi Li

{"title":"ReT-UAP: Rethinking Multi-Surrogate Universal Adversarial Perturbations on Vision-Language Pre-trained Models","authors":"Ziyu Wang, Jian Li, Yiming Du, Rui Ning, Jiang Li, Daniel Takabi, Lusi Li","doi":"10.1016/j.inffus.2026.104913","DOIUrl":"https://doi.org/10.1016/j.inffus.2026.104913","url":null,"abstract":"Universal adversarial perturbations (UAPs) on vision–language pre-trained (VLP) models are typically optimized with a non-targeted objective that repels each perturbed embedding from its own clean counterpart. Such an objective is defined relative to each surrogate: it specifies what to move away from, but not where to move. We observe that this property limits how well multi-surrogate training scales—when surrogates do not share a common embedding geometry, their repulsion directions are locally valid yet mutually uncoordinated, and the marginal value of consulting an additional surrogate diminishes. In this paper, we argue that the objective, rather than the selection policy, determines whether multi-surrogate training composes, and propose <ce:bold>ReT-UAP</ce:bold>. Each surrogate constructs paired cross-modal prototypes from a clean feature bank and derives soft bimodal targets via entropic optimal transport, scored primarily by semantic mismatch with hubness and cross-modal diversity as structural priors, so that every surrogate drives the perturbation toward a position determined by global cross-modal structure rather than away from an arbitrary origin. Modality-specific conditional generators are then optimized with a margin-aware dual-path contrastive objective with informed negatives. On top of this, an Upper Confidence Bound (UCB) policy with normalized rewards activates a single auxiliary surrogate per step, keeping the per-step cost close to single-surrogate training regardless of pool size. Targeted guidance makes the gradients composable; dynamic selection makes composing them affordable. Experiments on Flickr30K and MS-COCO across six VLP models spanning fused- and aligned-encoder families show that ReT-UAP substantially improves effectiveness on non-primary models over single-surrogate baselines, while remaining robust under common input-preprocessing defenses.","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"103 1","pages":""},"PeriodicalIF":18.6,"publicationDate":"2026-11-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149359135","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}

引用次数: 0

CGmini: High-order Consistent Graph Mining with Dual-level Cross-path Relation Enhancement CGmini：基于双级交叉路径关系增强的高阶一致性图挖掘

IF 18.6 1区计算机科学

Information Fusion Pub Date : 2026-11-02 DOI: 10.1016/j.inffus.2026.104916

Weihong Lin, Zhenghong Lin, Zhaoliang Chen, Yuhong Chen, Linmin Huang, Yiqing Shi, Shiping Wang

{"title":"CGmini: High-order Consistent Graph Mining with Dual-level Cross-path Relation Enhancement","authors":"Weihong Lin, Zhenghong Lin, Zhaoliang Chen, Yuhong Chen, Linmin Huang, Yiqing Shi, Shiping Wang","doi":"10.1016/j.inffus.2026.104916","DOIUrl":"https://doi.org/10.1016/j.inffus.2026.104916","url":null,"abstract":"Heterogeneous graph learning seeks to uncover meaningful patterns in graphs comprising a diverse spectrum of node and edge types, enabling effective representations for downstream tasks. However, existing methods are constrained by predefined path length, which restricts their ability to effectively model long-range dependencies extending beyond fixed scopes, ultimately diminishing the quality of node embeddings. To overcome this limitation, we propose a dual-level fusion framework that integrates insights from both topology and distance perspectives, thereby enhancing the extraction of comprehensive dependencies. For the observed topologies, the framework captures complex, multi-hop interactions by allowing nodes to traverse across different meta-paths, modeling dependencies that exceed the fixed-length limitation. Furthermore, it constructs global relations by leveraging attribute-driven distance correlations, offering more complete dependencies of the graph. In addition, we introduce a collaborative training mechanism ensuring a synergistic fusion of representations and enhancing the generalization capability of the model. Extensive experimental evaluations conducted on benchmark heterogeneous graph datasets validate the effectiveness of this approach in improving the performance of classification.","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"18 1","pages":""},"PeriodicalIF":18.6,"publicationDate":"2026-11-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149359134","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}

引用次数: 0

AIS-Vision Fusion for Robust Maritime Vessel Instance Segmentation under Visual Corruptions 基于ais -视觉融合的船舶实例分割

IF 18.6 1区计算机科学

Information Fusion Pub Date : 2026-10-28 DOI: 10.1016/j.inffus.2026.104909

Yunjia Wang, Zhongbi Luo, Zhenyi Wang, Kaizheng Wang, Holger Caesar, Jeroen Boydens, Davy Pissoort, Mathias Verbeke

{"title":"AIS-Vision Fusion for Robust Maritime Vessel Instance Segmentation under Visual Corruptions","authors":"Yunjia Wang, Zhongbi Luo, Zhenyi Wang, Kaizheng Wang, Holger Caesar, Jeroen Boydens, Davy Pissoort, Mathias Verbeke","doi":"10.1016/j.inffus.2026.104909","DOIUrl":"https://doi.org/10.1016/j.inffus.2026.104909","url":null,"abstract":"Robust visual perception is critical for maritime situational awareness across diverse waterborne environments, including inland waterways, particularly for safety-critical applications such as Autonomous Surface Vehicles (ASVs). However, vision-only models remain highly vulnerable to degradation from natural corruptions, such as adverse weather, noise, blur, and digital distortions. While multimodal fusion is a common strategy to enhance robustness, existing approaches typically rely on comparatively high-cost sensing modalities, such as LiDAR and marine-grade millimeter-wave radar, and largely overlook the potential of the widely available Automatic Identification System (AIS). Focusing on image-level perception, we propose a novel plug-and-play multimodal learning framework that leverages AIS as a spatially grounded, image-corruption-invariant prior to guide visual feature learning. Instead of treating AIS as a post-detection cue, we project AIS signals into image space and integrate them at the feature-learning stage via an adaptive gated fusion mechanism, enabling effective interaction between heterogeneous modalities without redesigning the underlying visual models. To support systematic evaluation, we introduce FVessel-InstAIS, the first synchronized AIS-vision dataset with instance segmentation annotations and camera calibration data, along with FVessel-InstAIS-C, a corruption benchmark covering 15 corruption types and multiple severity levels. Experiments demonstrate that the proposed framework consistently improves robustness across multiple base models under diverse corruptions, while maintaining competitive performance on clean data. Our findings provide empirical evidence that even sparse, noisy signals like AIS can effectively regularize visual learning when spatially aligned and adaptively fused with image content, offering a practical, low-cost pathway to more dependable maritime perception and insights for broader multimodal robustness.","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"6 1","pages":""},"PeriodicalIF":18.6,"publicationDate":"2026-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149292747","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}

引用次数: 0

C2C-GFL: Cluster-to-Class Fusion for Graph Few-Shot Learning C2C-GFL：聚类到类融合的图少镜头学习

IF 18.6 1区计算机科学

Information Fusion Pub Date : 2026-10-26 DOI: 10.1016/j.inffus.2026.104898

Haoran Yang, Rui Duan, Xin Guo, Rui Liu, Ke Liu

{"title":"C2C-GFL: Cluster-to-Class Fusion for Graph Few-Shot Learning","authors":"Haoran Yang, Rui Duan, Xin Guo, Rui Liu, Ke Liu","doi":"10.1016/j.inffus.2026.104898","DOIUrl":"https://doi.org/10.1016/j.inffus.2026.104898","url":null,"abstract":"Graph few-shot learning aims to classify nodes in graph-structured data when only a few labeled examples are available for each class. From an information fusion perspective, this task requires effective integration of node attributes, graph topology, sparse label evidence, and latent cluster structure. To mitigate the accuracy loss caused by label scarcity, several methods introduce deep clustering to provide auxiliary priors for downstream classification. However, most of these methods use clustering mainly as an intermediate source of pseudo-labels or auxiliary priors, so the structural and semantic partition information captured by clustering contributes to final prediction only indirectly. To address this gap, we propose C2C-GFL, a cluster-to-class fusion framework for graph few-shot learning that directly transforms label-aware cluster partitions into node classification results. Specifically, available labels are first incorporated into graph representation learning by bringing labeled nodes from the same class closer while separating those from different classes. An evidence-fused clustering state selection mechanism then integrates labeled-node concentration with graph-structure consistency and retains the highest-scoring state encountered during training. Finally, cluster-to-class alignment estimates cluster-class correspondence by fusing embedding-space prototype proximity with the empirical class composition of labeled nodes within each cluster. The resulting correspondence is used directly as the prediction mechanism for unlabeled nodes. Experiments on multiple benchmark datasets demonstrate that C2C-GFL achieves competitive performance improvements over representative graph few-shot learning methods under different few-shot settings.","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"18 1","pages":""},"PeriodicalIF":18.6,"publicationDate":"2026-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149292489","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}

引用次数: 0

Corrigendum to “Reducing causal attribution ambiguity in knowledge tracing via cognitive decoupling and adaptive fusion” [Information Fusion 137 (2027) 104648] “通过认知解耦和自适应融合减少知识追踪中的因果归因歧义”的更正[Information fusion 137 (2027) 104648]

IF 18.6 1区计算机科学

Information Fusion Pub Date : 2026-10-24 DOI: 10.1016/j.inffus.2026.104878

Yanqiang Wu, Youheng Bai, Zhiwei Huang, Yibo Lv, Mingliang Hou, Zitao Liu

引用次数: 0