Information Fusion最新文献_第5页

MSTFDN: An EEG-fNIRS multimodal spatial-temporal fusion decoding network for personalized multi-task scenarios MSTFDN：面向个性化多任务场景的EEG-fNIRS多模态时空融合解码网络

IF 15.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Information Fusion

Pub Date : 2026-07-01 Epub Date: 2026-01-23 DOI: 10.1016/j.inffus.2026.104187

Peng Ding , Liyong Yin , Zhengxuan Zhou , Yuwei Su , Minqian Zhang , Yingwei Li , Xiaoli Li

Multimodal information enables Brain-Computer Interface (BCI) systems to adapt to the differences in individual neural characteristics, overcoming the limitations of each modality. As a result, multimodal fusion technology that integrates non-invasive brain imaging techniques such as electroencephalogram (EEG) and functional near-infrared spectroscopy (fNIRS) has gained widespread attention. However, in the field of hybrid BCI, challenges remain in effectively integrating the heterogeneous information from these two modalities and improving the decoding accuracy and generalization across various task conditions. The core issue lies in the underutilization of each modality’s signal characteristics and the incomplete capture of the potential homogeneity of higher-order hybrid features. Therefore, we propose a novel EEG-fNIRS multimodal spatial-temporal fusion decoding network (MSTFDN). This network combines multi-scale temporal convolution of time series differences and spatial multi-head self-attention mechanism. MSTFDN consists of three core components, including the EEG branch, the fNIRS branch, the EEG-fNIRS fusion branch. A multi-dimensional loss function is constructed based on independent and hybrid space multi-head expression diversity, aiming to achieve high-precision decoding in small sample datasets under multi-task and multiple personalized experimental protocols. In experiments with four motor imagery (MI) and mental workload (MWL) tasks of two public datasets under three personalized experimental protocols, MSTFDN demonstrated state-of-the-art performance. The more comprehensive experimental protocols may establish a benchmark for model performance evaluation for future research in this field. Meanwhile, MSTFDN is also expected to become a new benchmark method for EEG-fNIRS hybrid BCI research.

多模态信息使脑机接口（BCI）系统能够适应个体神经特征的差异，克服每种模态的局限性。因此，结合脑电图（EEG）和功能近红外光谱（fNIRS）等非侵入性脑成像技术的多模态融合技术得到了广泛的关注。然而，在混合脑机接口领域，如何有效地整合这两种模式的异构信息，提高不同任务条件下的解码精度和泛化程度仍然是一个挑战。核心问题在于每种模态的信号特征未被充分利用，高阶混合特征的潜在同质性未被完全捕获。为此，我们提出了一种新的EEG-fNIRS多模态时空融合解码网络（MSTFDN）。该网络结合了时间序列差异的多尺度时间卷积和空间多头自注意机制。MSTFDN由EEG分支、fNIRS分支、EEG-fNIRS融合分支三个核心部分组成。基于独立和混合空间多头表达多样性构建了多维损失函数，旨在实现多任务、多个性化实验方案下小样本数据集的高精度解码。在三个个性化实验方案下，MSTFDN在两个公共数据集的四个运动意象（MI）和精神负荷（MWL）任务中表现出了最先进的性能。更全面的实验方案可以为未来该领域的研究建立模型性能评价的基准。同时，MSTFDN也有望成为EEG-fNIRS混合脑接口研究的新标杆方法。

{"title":"MSTFDN: An EEG-fNIRS multimodal spatial-temporal fusion decoding network for personalized multi-task scenarios","authors":"Peng Ding , Liyong Yin , Zhengxuan Zhou , Yuwei Su , Minqian Zhang , Yingwei Li , Xiaoli Li","doi":"10.1016/j.inffus.2026.104187","DOIUrl":"10.1016/j.inffus.2026.104187","url":null,"abstract":"<div><div>Multimodal information enables Brain-Computer Interface (BCI) systems to adapt to the differences in individual neural characteristics, overcoming the limitations of each modality. As a result, multimodal fusion technology that integrates non-invasive brain imaging techniques such as electroencephalogram (EEG) and functional near-infrared spectroscopy (fNIRS) has gained widespread attention. However, in the field of hybrid BCI, challenges remain in effectively integrating the heterogeneous information from these two modalities and improving the decoding accuracy and generalization across various task conditions. The core issue lies in the underutilization of each modality’s signal characteristics and the incomplete capture of the potential homogeneity of higher-order hybrid features. Therefore, we propose a novel EEG-fNIRS multimodal spatial-temporal fusion decoding network (MSTFDN). This network combines multi-scale temporal convolution of time series differences and spatial multi-head self-attention mechanism. MSTFDN consists of three core components, including the EEG branch, the fNIRS branch, the EEG-fNIRS fusion branch. A multi-dimensional loss function is constructed based on independent and hybrid space multi-head expression diversity, aiming to achieve high-precision decoding in small sample datasets under multi-task and multiple personalized experimental protocols. In experiments with four motor imagery (MI) and mental workload (MWL) tasks of two public datasets under three personalized experimental protocols, MSTFDN demonstrated state-of-the-art performance. The more comprehensive experimental protocols may establish a benchmark for model performance evaluation for future research in this field. Meanwhile, MSTFDN is also expected to become a new benchmark method for EEG-fNIRS hybrid BCI research.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"131 ","pages":"Article 104187"},"PeriodicalIF":15.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146033284","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

Speech emotion recognition: A systematic mega-review of techniques and pipelines 语音情感识别：技术和管道的系统综述

IF 15.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Information Fusion

Pub Date : 2026-07-01 Epub Date: 2026-01-18 DOI: 10.1016/j.inffus.2026.104161

Adil Chakhtouna, Sara Sekkate, Abdellah Adib

Speech Emotion Recognition (SER) is a rapidly evolving research field that aims to enable machines to automatically identify human emotions from vocal signals. This systematic review presents a comprehensive and structured synthesis of the SER literature, focusing on eleven key research questions that span the theoretical foundations, signal processing pipeline, and methodological advancements in the field. Unlike prior surveys, this review unifies both foundational and state-of-the-art insights across all stages of the SER pipeline through a question-driven structure, offering a clear road-map for both new and experienced researchers in the SER community. We first explore the psychological and computational modeling of emotions, followed by a detailed examination of the different modalities for emotion expression, with a particular emphasis on speech. The review highlights the most widely used emotional speech databases, common pre-processing techniques, and the diverse set of handcrafted and learned features employed in SER. We compare traditional machine learning approaches with recent deep learning models, emphasizing their respective strengths, limitations, and application contexts. Special attention is given to the recent shift toward self-supervised learning (SSL) models such as Wav2Vec2 and HuBERT, which have redefined the state-of-the-art in speech-based representation learning. Special attention is given to evaluation metrics, benchmarking strategies, and real-world deployment challenges, including issues of speaker-independence and environmental variability. The review concludes by identifying key limitations across the literature and articulating future research directions necessary for developing reliable, scalable, and context-aware emotion-aware systems. Overall, this work serves as a central reference for advancing SER research and practical deployment in real-world environments.

语音情感识别（SER）是一个快速发展的研究领域，旨在使机器能够从声音信号中自动识别人类的情感。本系统综述对SER文献进行了全面和结构化的综合，重点关注11个关键研究问题，这些问题涵盖了该领域的理论基础、信号处理管道和方法进展。与之前的调查不同，该综述通过一个问题驱动的结构，统一了SER管道所有阶段的基础和最新的见解，为SER社区的新研究人员和经验丰富的研究人员提供了清晰的路线图。我们首先探索情绪的心理和计算建模，然后详细检查情绪表达的不同方式，特别强调语言。这篇综述强调了最广泛使用的情绪语音数据库，常见的预处理技术，以及SER中使用的各种手工和学习特征。我们将传统的机器学习方法与最近的深度学习模型进行比较，强调它们各自的优势、局限性和应用环境。特别关注最近向自我监督学习（SSL）模型的转变，如Wav2Vec2和HuBERT，它们重新定义了基于语音的表示学习的最新技术。特别关注评估指标、基准策略和现实世界的部署挑战，包括说话者独立性和环境可变性问题。本文总结了文献中的关键限制，并阐明了开发可靠、可扩展和上下文感知的情感感知系统所需的未来研究方向。总的来说，这项工作是推进SER研究和在现实环境中实际部署的核心参考。

{"title":"Speech emotion recognition: A systematic mega-review of techniques and pipelines","authors":"Adil Chakhtouna, Sara Sekkate, Abdellah Adib","doi":"10.1016/j.inffus.2026.104161","DOIUrl":"10.1016/j.inffus.2026.104161","url":null,"abstract":"<div><div>Speech Emotion Recognition (SER) is a rapidly evolving research field that aims to enable machines to automatically identify human emotions from vocal signals. This systematic review presents a comprehensive and structured synthesis of the SER literature, focusing on eleven key research questions that span the theoretical foundations, signal processing pipeline, and methodological advancements in the field. Unlike prior surveys, this review unifies both foundational and state-of-the-art insights across all stages of the SER pipeline through a question-driven structure, offering a clear road-map for both new and experienced researchers in the SER community. We first explore the psychological and computational modeling of emotions, followed by a detailed examination of the different modalities for emotion expression, with a particular emphasis on speech. The review highlights the most widely used emotional speech databases, common pre-processing techniques, and the diverse set of handcrafted and learned features employed in SER. We compare traditional machine learning approaches with recent deep learning models, emphasizing their respective strengths, limitations, and application contexts. Special attention is given to the recent shift toward self-supervised learning (SSL) models such as Wav2Vec2 and HuBERT, which have redefined the state-of-the-art in speech-based representation learning. Special attention is given to evaluation metrics, benchmarking strategies, and real-world deployment challenges, including issues of speaker-independence and environmental variability. The review concludes by identifying key limitations across the literature and articulating future research directions necessary for developing reliable, scalable, and context-aware emotion-aware systems. Overall, this work serves as a central reference for advancing SER research and practical deployment in real-world environments.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"131 ","pages":"Article 104161"},"PeriodicalIF":15.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145995202","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

Fusion of quantum computing with smart agriculture: A systematic review of methods, implementation, applications, and challenges 量子计算与智慧农业的融合：方法、实现、应用和挑战的系统回顾

IF 15.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Information Fusion

Pub Date : 2026-07-01 Epub Date: 2026-01-16 DOI: 10.1016/j.inffus.2026.104159

Sumit Kumar , Shashank Sheshar Singh , Gourav Bathla , Swati Sharma , Manisha Panjeta

The growing global population and the severity of environmental issues are driving the agriculture sector to adopt innovative technological advances for sustainable food production. Classical computing approaches frequently struggle with the volume and complexity of agricultural data when performing tasks such as crop yield prediction, disease detection, soil analysis, and weather forecasting. This Systematic Literature Review (SLR) provides an in-depth analysis of the evolving significance of quantum computing in smart agriculture. Quantum algorithms have the potential to reduce computational complexity and create novel data representation methods for high-dimensional challenges by leveraging quantum mechanics principles such as superposition and entanglement. This paper employs a structured research methodology based on eight specific research questions to comprehensively investigate over 100 peer-reviewed studies on quantum computing and smart agriculture published between 2012 and 2025. It demonstrates the effectiveness of Quantum Machine Learning (QML), quantum optimization, and hybrid quantum-classical models in various agricultural applications. The survey examines real-world implementations and compares existing quantum initiatives to classical benchmarks for the classification and prediction tasks. The presented work identifies challenges and limitations of current quantum approaches. The paper outlines directions for future work, including the accessibility of quantum hardware and the development of domain-specific algorithms. To the best of our knowledge, this is the first research question-driven SLR that provides an in-depth analysis of how quantum computing can be applied in agricultural applications.

全球人口的不断增长和环境问题的严重程度正在推动农业部门采用创新的技术进步来实现可持续的粮食生产。在执行诸如作物产量预测、疾病检测、土壤分析和天气预报等任务时，经典计算方法经常与农业数据的数量和复杂性作斗争。本系统文献综述（SLR）深入分析了量子计算在智能农业中的发展意义。量子算法有可能通过利用量子力学原理（如叠加和纠缠）来降低计算复杂性，并为高维挑战创造新的数据表示方法。本文采用基于8个具体研究问题的结构化研究方法，全面调查了2012年至2025年间发表的100多篇同行评审的量子计算和智慧农业研究。它展示了量子机器学习（QML），量子优化和混合量子经典模型在各种农业应用中的有效性。该调查考察了现实世界的实现，并将现有的量子计划与分类和预测任务的经典基准进行了比较。提出的工作确定了当前量子方法的挑战和局限性。本文概述了未来工作的方向，包括量子硬件的可访问性和特定领域算法的发展。据我们所知，这是第一个研究问题驱动的单反，它提供了量子计算如何应用于农业应用的深入分析。

{"title":"Fusion of quantum computing with smart agriculture: A systematic review of methods, implementation, applications, and challenges","authors":"Sumit Kumar , Shashank Sheshar Singh , Gourav Bathla , Swati Sharma , Manisha Panjeta","doi":"10.1016/j.inffus.2026.104159","DOIUrl":"10.1016/j.inffus.2026.104159","url":null,"abstract":"<div><div>The growing global population and the severity of environmental issues are driving the agriculture sector to adopt innovative technological advances for sustainable food production. Classical computing approaches frequently struggle with the volume and complexity of agricultural data when performing tasks such as crop yield prediction, disease detection, soil analysis, and weather forecasting. This Systematic Literature Review (SLR) provides an in-depth analysis of the evolving significance of quantum computing in smart agriculture. Quantum algorithms have the potential to reduce computational complexity and create novel data representation methods for high-dimensional challenges by leveraging quantum mechanics principles such as superposition and entanglement. This paper employs a structured research methodology based on eight specific research questions to comprehensively investigate over 100 peer-reviewed studies on quantum computing and smart agriculture published between 2012 and 2025. It demonstrates the effectiveness of Quantum Machine Learning (QML), quantum optimization, and hybrid quantum-classical models in various agricultural applications. The survey examines real-world implementations and compares existing quantum initiatives to classical benchmarks for the classification and prediction tasks. The presented work identifies challenges and limitations of current quantum approaches. The paper outlines directions for future work, including the accessibility of quantum hardware and the development of domain-specific algorithms. To the best of our knowledge, this is the first research question-driven SLR that provides an in-depth analysis of how quantum computing can be applied in agricultural applications.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"131 ","pages":"Article 104159"},"PeriodicalIF":15.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145995251","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

Controlled subspace fusion for language model continual learning 语言模型持续学习控制子空间融合

IF 15.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Information Fusion

Pub Date : 2026-07-01 Epub Date: 2026-01-24 DOI: 10.1016/j.inffus.2026.104184

Xingcan Bao , Jianzhou Feng , Yiru Huo , Huaxiao Qiu , Haoran Yu , Shenyuan Ren , Jiadong Ren

Large language models (LLMs) have demonstrated remarkable performance across diverse natural language processing tasks. However, they still face significant challenges in multi-task continual learning, particularly in dynamic environments where tasks evolve sequentially and resources are constrained. Existing approaches typically learn separate adapter modules for each task, leading to a linear increase in parameters as tasks accumulate and thus hindering scalability and deployment efficiency. In this paper, we propose Controlled Subspace Fusion (CSF), a rehearsal-free and task-agnostic continual learning framework for language models that integrates knowledge across tasks while preventing parameter explosion. CSF introduces a shared low-rank projection subspace to provide a unified representational foundation, thereby enhancing consistency and facilitating cross-task knowledge transfer. In addition, we design an incremental subspace fusion mechanism that adaptively merges new task adapters with previously fused representations, while suppressing redundant parameter growth. As a result, the framework achieves scalable and robust knowledge fusion across sequential tasks. We evaluate CSF on mainstream architectures, including LLaMA and T5, across model scales ranging from 220M to 13B parameters. Experimental results on continual learning benchmarks demonstrate that CSF not only achieves superior average accuracy and parameter efficiency compared to existing approaches, but also provides a scalable and deployment-friendly solution that supports efficient knowledge fusion.

大型语言模型（llm）在不同的自然语言处理任务中表现出了显著的性能。然而，他们在多任务持续学习中仍然面临着重大挑战，特别是在任务顺序演变和资源受限的动态环境中。现有的方法通常为每个任务学习单独的适配器模块，导致参数随着任务的积累呈线性增长，从而阻碍了可伸缩性和部署效率。在本文中，我们提出了受控子空间融合（CSF），这是一种无需预演且与任务无关的语言模型持续学习框架，可在防止参数爆炸的同时集成跨任务的知识。CSF引入共享的低秩投影子空间，提供统一的表示基础，从而增强一致性，促进跨任务知识转移。此外，我们设计了一种增量子空间融合机制，该机制自适应地将新的任务适配器与先前融合的表示合并，同时抑制冗余参数的增长。因此，该框架实现了跨顺序任务的可伸缩和健壮的知识融合。我们在主流架构上评估CSF，包括LLaMA和T5，模型尺度从220M到13B参数。持续学习基准的实验结果表明，与现有方法相比，CSF不仅具有更高的平均精度和参数效率，而且还提供了可扩展和部署友好的解决方案，支持高效的知识融合。

{"title":"Controlled subspace fusion for language model continual learning","authors":"Xingcan Bao , Jianzhou Feng , Yiru Huo , Huaxiao Qiu , Haoran Yu , Shenyuan Ren , Jiadong Ren","doi":"10.1016/j.inffus.2026.104184","DOIUrl":"10.1016/j.inffus.2026.104184","url":null,"abstract":"<div><div>Large language models (LLMs) have demonstrated remarkable performance across diverse natural language processing tasks. However, they still face significant challenges in multi-task continual learning, particularly in dynamic environments where tasks evolve sequentially and resources are constrained. Existing approaches typically learn separate adapter modules for each task, leading to a linear increase in parameters as tasks accumulate and thus hindering scalability and deployment efficiency. In this paper, we propose Controlled Subspace Fusion (CSF), a rehearsal-free and task-agnostic continual learning framework for language models that integrates knowledge across tasks while preventing parameter explosion. CSF introduces a shared low-rank projection subspace to provide a unified representational foundation, thereby enhancing consistency and facilitating cross-task knowledge transfer. In addition, we design an incremental subspace fusion mechanism that adaptively merges new task adapters with previously fused representations, while suppressing redundant parameter growth. As a result, the framework achieves scalable and robust knowledge fusion across sequential tasks. We evaluate CSF on mainstream architectures, including LLaMA and T5, across model scales ranging from 220M to 13B parameters. Experimental results on continual learning benchmarks demonstrate that CSF not only achieves superior average accuracy and parameter efficiency compared to existing approaches, but also provides a scalable and deployment-friendly solution that supports efficient knowledge fusion.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"131 ","pages":"Article 104184"},"PeriodicalIF":15.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146047984","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

SU-RMT: Toward bridging semantic representation and structural detail modeling for medical image segmentation 面向医学图像分割的语义表示和结构细节建模

IF 15.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Information Fusion

Pub Date : 2026-07-01 Epub Date: 2026-01-26 DOI: 10.1016/j.inffus.2026.104182

Peibo Song , Zihao Wang , Jinshuo Zhang , Shujun Fu , Yunfeng Zhang , Wei Wu , Fangxun Bao

Accurate medical image segmentation requires models that capture high-level semantics while preserving fine-grained structural details, due to anatomical heterogeneity and subtle textures in clinical scenarios. However, existing U-shaped networks usually lack a unified perspective to reconcile semantic representation with structural detail. To this end, we present SU-RMT, a U-shaped network that embodies this unified perspective by redesigning the encoder, bottleneck, and skip connection. The encoder employs the Dynamic Spatial Attention (DySA) mechanism to capture global context with spatial priors. The bottleneck introduces a Hybrid Spectral Adaptive (HSA) module to transform abstract semantics into structure-aware features. The first skip connection incorporates a Frequency-Fused (F²) block to enhance boundary details without amplifying noise. Across several medical image segmentation tasks, SU-RMT demonstrates strong performance. The code is at the link.

由于临床场景中的解剖异质性和微妙纹理，准确的医学图像分割需要模型捕获高级语义，同时保留细粒度的结构细节。然而，现有的u型网络通常缺乏统一的视角来协调语义表示和结构细节。为此，我们提出了SU-RMT，这是一个u形网络，通过重新设计编码器、瓶颈和跳过连接来体现这种统一的观点。编码器采用动态空间注意（DySA）机制捕捉具有空间先验的全局上下文。瓶颈引入了混合光谱自适应（HSA）模块，将抽象语义转换为结构感知特征。第一跳过连接包含频率融合（F2）块，以增强边界细节而不会放大噪声。在多个医学图像分割任务中，SU-RMT显示出强大的性能。代码在链接处。

{"title":"SU-RMT: Toward bridging semantic representation and structural detail modeling for medical image segmentation","authors":"Peibo Song , Zihao Wang , Jinshuo Zhang , Shujun Fu , Yunfeng Zhang , Wei Wu , Fangxun Bao","doi":"10.1016/j.inffus.2026.104182","DOIUrl":"10.1016/j.inffus.2026.104182","url":null,"abstract":"<div><div>Accurate medical image segmentation requires models that capture high-level semantics while preserving fine-grained structural details, due to anatomical heterogeneity and subtle textures in clinical scenarios. However, existing U-shaped networks usually lack a unified perspective to reconcile semantic representation with structural detail. To this end, we present <strong>SU-RMT</strong>, a U-shaped network that embodies this unified perspective by redesigning the encoder, bottleneck, and skip connection. The encoder employs the <strong>Dy</strong>namic <strong>S</strong>patial <strong>A</strong>ttention <strong>(DySA)</strong> mechanism to capture global context with spatial priors. The bottleneck introduces a <strong>H</strong>ybrid <strong>S</strong>pectral <strong>A</strong>daptive <strong>(HSA)</strong> module to transform abstract semantics into structure-aware features. The first skip connection incorporates a <strong>F</strong>requency-<strong>F</strong>used <strong>(F</strong><sup>2</sup><strong>)</strong> block to enhance boundary details without amplifying noise. Across several medical image segmentation tasks, SU-RMT demonstrates strong performance. The code is at the <span><span>link</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"131 ","pages":"Article 104182"},"PeriodicalIF":15.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146048053","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

Unsupervised multimodal graph completion networks with multi-level contrastiveness for modality-missing conversation understanding 基于多级对比的无监督多模态图补全网络用于情态缺失会话理解

IF 15.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Information Fusion

Pub Date : 2026-07-01 Epub Date: 2026-01-29 DOI: 10.1016/j.inffus.2026.104197

Sichao Fu , Songren Peng , Bin Zou , Xiao-Yuan Jing , Wei Yu , Qinmu Peng , Xinge You

Multimodal conversation understanding has received increasing research interest in recent years, which aims to integrate multimodal conversation information to improve the accuracy of computer understanding of user intentions. However, the existing multimodal conversation understanding methods often suffer from a conversation modality missing challenge, which seriously damages their superior performance. Recently emerged imputation-based incomplete multimodal learning (I²ML) provides an effective solution, which aims to reconstruct the missing modality features under the supervision of a downstream task. Such reliance on labels causes both the bias of the reconstructed modality features and the limitation of their scope of application. Besides, these proposed I²ML methods independently consider the missing modality features reconstruction process between different utterances, which further leads to a specific utterance over-reliance (model sub-optimal) issue. To address the above-mentioned issues, a more general unsupervised I²ML is proposed to effectively improve the performance of the modality-missing conversation understanding (M²CU) task, termed unsupervised multimodal graph completion networks (UMGCN). Specifically, to improve the accuracy of each reconstructed modality feature, an effective missing modality recovery module is designed to enhance the information interaction process between different utterances for generating robust missing modality recovery features. Then, a multi-level graph contrastive loss on the cross-structure and cross-view level is proposed to learn utterance-general conversation representations by maximizing the mutual information between the same conversation representations across different structures and views. Finally, the learned utterance-general conversation representations can be applied to arbitrary M²CU tasks. Extensive experiments on four datasets, seven missing rates and two M²CU tasks show that our proposed UMGCN outperforms the existing incomplete multimodal learning methods.

多模态会话理解近年来受到越来越多的研究兴趣，其目的是整合多模态会话信息以提高计算机对用户意图理解的准确性。然而，现有的多模态会话理解方法往往存在会话模态缺失的问题，严重影响了多模态会话理解方法的良好性能。最近出现的基于输入的不完全多模态学习（I2ML）提供了一种有效的解决方案，其目的是在下游任务的监督下重建缺失的模态特征。这种对标签的依赖既造成重构模态特征的偏倚，又限制了它们的适用范围。此外，这些I2ML方法独立考虑了不同话语之间缺失的情态特征重建过程，从而导致特定的话语过度依赖（模型次优）问题。为了解决上述问题，提出了一种更通用的无监督I2ML，以有效地提高模态缺失会话理解（M2CU）任务的性能，称为无监督多模态图完成网络（UMGCN）。具体而言，为了提高各重构情态特征的准确性，设计了有效的缺失情态恢复模块，增强不同话语之间的信息交互过程，生成鲁棒的缺失情态恢复特征。然后，提出了一种跨结构和跨视图层次的多层次图对比损失，通过最大化跨不同结构和视图的相同会话表示之间的互信息来学习话语-通用会话表示。最后，学习到的话语-一般会话表示可以应用于任意的M2CU任务。在4个数据集、7个缺失率和2个M2CU任务上的大量实验表明，我们提出的UMGCN优于现有的不完全多模态学习方法。

{"title":"Unsupervised multimodal graph completion networks with multi-level contrastiveness for modality-missing conversation understanding","authors":"Sichao Fu , Songren Peng , Bin Zou , Xiao-Yuan Jing , Wei Yu , Qinmu Peng , Xinge You","doi":"10.1016/j.inffus.2026.104197","DOIUrl":"10.1016/j.inffus.2026.104197","url":null,"abstract":"<div><div>Multimodal conversation understanding has received increasing research interest in recent years, which aims to integrate multimodal conversation information to improve the accuracy of computer understanding of user intentions. However, the existing multimodal conversation understanding methods often suffer from a conversation modality missing challenge, which seriously damages their superior performance. Recently emerged imputation-based incomplete multimodal learning (I<sup>2</sup>ML) provides an effective solution, which aims to reconstruct the missing modality features under the supervision of a downstream task. Such reliance on labels causes both the bias of the reconstructed modality features and the limitation of their scope of application. Besides, these proposed I<sup>2</sup>ML methods independently consider the missing modality features reconstruction process between different utterances, which further leads to a specific utterance over-reliance (model sub-optimal) issue. To address the above-mentioned issues, a more general unsupervised I<sup>2</sup>ML is proposed to effectively improve the performance of the modality-missing conversation understanding (M<sup>2</sup>CU) task, termed unsupervised multimodal graph completion networks (UMGCN). Specifically, to improve the accuracy of each reconstructed modality feature, an effective missing modality recovery module is designed to enhance the information interaction process between different utterances for generating robust missing modality recovery features. Then, a multi-level graph contrastive loss on the cross-structure and cross-view level is proposed to learn utterance-general conversation representations by maximizing the mutual information between the same conversation representations across different structures and views. Finally, the learned utterance-general conversation representations can be applied to arbitrary M<sup>2</sup>CU tasks. Extensive experiments on four datasets, seven missing rates and two M<sup>2</sup>CU tasks show that our proposed UMGCN outperforms the existing incomplete multimodal learning methods.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"131 ","pages":"Article 104197"},"PeriodicalIF":15.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146071716","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

MRFNet: Multi-reference fusion for image deblurring MRFNet：多参考融合图像去模糊

IF 15.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Information Fusion

Pub Date : 2026-07-01 Epub Date: 2026-01-22 DOI: 10.1016/j.inffus.2026.104169

Tingrui Guo , Chi Xu , Kaifeng Tang , Hao Qian

Motion blur is a persistent challenge in visual data processing. While single-image deblurring methods have made significant progress, using multiple reference images from the same scene for deblurring remains an overlooked problem. Existing methods struggle to integrate information from multiple reference images with differences in lighting, color, and perspective. Herein, we propose a novel framework MRFNet which leverages any number of discontinuous reference images for deblurring. The framework consists of two key components: (1) the Offset Fusion Module (OFM) guided by dense matching, which aggregates features from discontinuous reference images through high-frequency detail enhancement and permutation-invariant units; and (2) the Deformable Enrichment Module (DEM), which refines misaligned features using deformable convolutions for precise detail recovery. Quantitative and qualitative evaluations on synthetic and real-world datasets show that the proposed method outperforms state-of-the-art deblurring approaches. Additionally, a new real-world dataset is provided to fill the gap in evaluating discontinuous reference problems.

运动模糊一直是视觉数据处理中的难题。虽然单幅图像去模糊方法已经取得了重大进展，但使用来自同一场景的多幅参考图像进行去模糊仍然是一个被忽视的问题。现有的方法很难整合来自多个参考图像的信息，这些图像在光照、颜色和视角上存在差异。在此，我们提出了一个新的框架MRFNet利用任意数量的不连续参考图像去模糊。该框架由两个关键部分组成：(1)偏移融合模块（OFM）以密集匹配为指导，通过高频细节增强和置换不变单元聚合不连续参考图像的特征；(2)可变形浓缩模块（DEM），它使用可变形卷积来细化不对齐的特征，以实现精确的细节恢复。对合成和真实世界数据集的定量和定性评估表明，所提出的方法优于最先进的去模糊方法。此外，还提供了一个新的真实数据集来填补评估不连续参考问题的空白。

{"title":"MRFNet: Multi-reference fusion for image deblurring","authors":"Tingrui Guo , Chi Xu , Kaifeng Tang , Hao Qian","doi":"10.1016/j.inffus.2026.104169","DOIUrl":"10.1016/j.inffus.2026.104169","url":null,"abstract":"<div><div>Motion blur is a persistent challenge in visual data processing. While single-image deblurring methods have made significant progress, using multiple reference images from the same scene for deblurring remains an overlooked problem. Existing methods struggle to integrate information from multiple reference images with differences in lighting, color, and perspective. Herein, we propose a novel framework MRFNet which leverages any number of discontinuous reference images for deblurring. The framework consists of two key components: (1) the Offset Fusion Module (OFM) guided by dense matching, which aggregates features from discontinuous reference images through high-frequency detail enhancement and permutation-invariant units; and (2) the Deformable Enrichment Module (DEM), which refines misaligned features using deformable convolutions for precise detail recovery. Quantitative and qualitative evaluations on synthetic and real-world datasets show that the proposed method outperforms state-of-the-art deblurring approaches. Additionally, a new real-world dataset is provided to fill the gap in evaluating discontinuous reference problems.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"131 ","pages":"Article 104169"},"PeriodicalIF":15.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146033291","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

A two-stage learning network for PVINS modeling and fusion estimation in challenging environments 挑战性环境下PVINS建模与融合估计的两阶段学习网络

IF 15.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Information Fusion

Pub Date : 2026-07-01 Epub Date: 2026-01-30 DOI: 10.1016/j.inffus.2026.104192

Xuanyu Wu , Jiankai Yin , Jian Yang , Xin Liu , Wenshuo Li , Lei Guo

In the polarization-based visual-inertial navigation system (PVINS), information from polarization sensor (PS) and visual-inertial navigation system (VINS) is fused to enable position and attitude estimation, thereby offering an effective solution for autonomous navigation in global navigation satellite system (GNSS)-denied environments. However, under challenging conditions such as complex weather, the state-space model of PVINS becomes susceptible to uncertain model error, limiting the accuracy and adaptability of the system. To address this issue, we propose a tightly coupled PVINS integration scheme based on a two-stage learning network, which consists of model error compensation and adaptive Kalman gain learning. In the first stage, a deep neural network with a shared-weight architecture is designed to learn and compensate for the state-space model error, thereby reducing network complexity and enabling more precise system modeling. In the second stage, to improve fusion accuracy of PVINS, a Kalman gain learning network (KGLN)-based intelligent fusion method is proposed. This approach enables the adaptive learning of Kalman gains, circumventing the dependency of the system on knowledge of the noise statistics. Finally, the performance of the system is verified through the semi-physical simulation and flight test. The experimental results confirm that the proposed method outperforms conventional PVINS in terms of both position and heading estimation.

基于极化的视觉惯性导航系统（PVINS）将极化传感器（PS）和视觉惯性导航系统（VINS）的信息融合在一起，实现位置和姿态估计，为全球导航卫星系统（GNSS）拒绝环境下的自主导航提供了有效的解决方案。然而，在复杂天气等具有挑战性的条件下，PVINS的状态空间模型容易受到不确定模型误差的影响，限制了系统的准确性和自适应性。为了解决这一问题，我们提出了一种基于两阶段学习网络的紧密耦合PVINS集成方案，该方案由模型误差补偿和自适应卡尔曼增益学习组成。在第一阶段，设计一个具有共享权重架构的深度神经网络来学习和补偿状态空间模型误差，从而降低网络复杂性，实现更精确的系统建模。第二阶段，为了提高PVINS的融合精度，提出了一种基于卡尔曼增益学习网络（KGLN）的智能融合方法。这种方法可以实现卡尔曼增益的自适应学习，避免了系统对噪声统计知识的依赖。最后，通过半实物仿真和飞行试验验证了系统的性能。实验结果表明，该方法在位置和航向估计方面都优于传统的PVINS。

{"title":"A two-stage learning network for PVINS modeling and fusion estimation in challenging environments","authors":"Xuanyu Wu , Jiankai Yin , Jian Yang , Xin Liu , Wenshuo Li , Lei Guo","doi":"10.1016/j.inffus.2026.104192","DOIUrl":"10.1016/j.inffus.2026.104192","url":null,"abstract":"<div><div>In the polarization-based visual-inertial navigation system (PVINS), information from polarization sensor (PS) and visual-inertial navigation system (VINS) is fused to enable position and attitude estimation, thereby offering an effective solution for autonomous navigation in global navigation satellite system (GNSS)-denied environments. However, under challenging conditions such as complex weather, the state-space model of PVINS becomes susceptible to uncertain model error, limiting the accuracy and adaptability of the system. To address this issue, we propose a tightly coupled PVINS integration scheme based on a two-stage learning network, which consists of model error compensation and adaptive Kalman gain learning. In the first stage, a deep neural network with a shared-weight architecture is designed to learn and compensate for the state-space model error, thereby reducing network complexity and enabling more precise system modeling. In the second stage, to improve fusion accuracy of PVINS, a Kalman gain learning network (KGLN)-based intelligent fusion method is proposed. This approach enables the adaptive learning of Kalman gains, circumventing the dependency of the system on knowledge of the noise statistics. Finally, the performance of the system is verified through the semi-physical simulation and flight test. The experimental results confirm that the proposed method outperforms conventional PVINS in terms of both position and heading estimation.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"131 ","pages":"Article 104192"},"PeriodicalIF":15.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146089493","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

Validity-aware context modeling for gradient-guided image inpainting 基于有效性感知的梯度引导图像绘制上下文建模

IF 15.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Information Fusion

Pub Date : 2026-07-01 Epub Date: 2026-01-19 DOI: 10.1016/j.inffus.2026.104162

Wuzhen Shi , Wu Yang , Zhihao Wu , Yang Wen

Existing prior-guided image inpainting methods show state-of-the-art performance. But their prior extraction is computationally expensive and unstable in accuracy. Besides, most of them only focus on the structure guidance, which hardly facilitates the repair of realistic textures. Inspired by the fact that gradient maps are easy to extract and reflect both image structure and fine texture details, this paper proposes a gradient-guided network for image inpainting, which first uses the gradient context information and multi-level image compensation features to repair the gradient, and then uses the repaired gradient features to guide the generation of realistic image. A gradient-driven attention (GDA) module is introduced for efficient prior guidance. Additionally, a context validity-aware (CVA) module is proposed for progressively filling hole regions of images, which accurately utilizes both local and contextual information for image inpainting via validity-aware measurements. Furthermore, by artificially manipulating the generation of the gradient map, our gradient-guided image inpainting method enables user-guided image editing, which effectively increases the diversity of image generation and enhances the flexibility of image editing. Experiments on benchmark datasets show that the proposed method outperforms the state-of-the-art methods. Extensive ablation experiments are also conducted to demonstrate the effectiveness of each module.

现有的先验引导图像绘制方法显示了最先进的性能。但它们的先验提取计算成本高，精度不稳定。此外，它们大多只注重结构引导，难以实现逼真纹理的修复。基于梯度图易于提取和反映图像结构和精细纹理细节的特点，本文提出了一种用于图像补图的梯度引导网络，该网络首先利用梯度上下文信息和多级图像补偿特征对梯度进行修复，然后利用修复后的梯度特征指导生成逼真的图像。引入梯度驱动注意力（GDA）模块，实现有效的事前引导。此外，提出了一种上下文有效性感知（CVA）模块，用于逐步填充图像的空洞区域，该模块通过有效性感知测量准确地利用局部和上下文信息进行图像绘制。此外，我们的梯度引导图像绘制方法通过人为操纵梯度图的生成，实现用户引导的图像编辑，有效地增加了图像生成的多样性，增强了图像编辑的灵活性。在基准数据集上的实验表明，该方法优于现有的方法。为了验证每个模块的有效性，还进行了大量的烧蚀实验。

{"title":"Validity-aware context modeling for gradient-guided image inpainting","authors":"Wuzhen Shi , Wu Yang , Zhihao Wu , Yang Wen","doi":"10.1016/j.inffus.2026.104162","DOIUrl":"10.1016/j.inffus.2026.104162","url":null,"abstract":"<div><div>Existing prior-guided image inpainting methods show state-of-the-art performance. But their prior extraction is computationally expensive and unstable in accuracy. Besides, most of them only focus on the structure guidance, which hardly facilitates the repair of realistic textures. Inspired by the fact that gradient maps are easy to extract and reflect both image structure and fine texture details, this paper proposes a gradient-guided network for image inpainting, which first uses the gradient context information and multi-level image compensation features to repair the gradient, and then uses the repaired gradient features to guide the generation of realistic image. A gradient-driven attention (GDA) module is introduced for efficient prior guidance. Additionally, a context validity-aware (CVA) module is proposed for progressively filling hole regions of images, which accurately utilizes both local and contextual information for image inpainting via validity-aware measurements. Furthermore, by artificially manipulating the generation of the gradient map, our gradient-guided image inpainting method enables user-guided image editing, which effectively increases the diversity of image generation and enhances the flexibility of image editing. Experiments on benchmark datasets show that the proposed method outperforms the state-of-the-art methods. Extensive ablation experiments are also conducted to demonstrate the effectiveness of each module.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"131 ","pages":"Article 104162"},"PeriodicalIF":15.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146000897","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

A novel knowledge distillation method for graph neural networks with gradient mapping and fusion 基于梯度映射和融合的图神经网络知识提取方法

IF 15.5 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Information Fusion

Pub Date : 2026-07-01 Epub Date: 2026-01-20 DOI: 10.1016/j.inffus.2026.104163

Kang Liu , Shunzhi Yang , Chang-Dong Wang , Yunwen Chen , Zhenhua Huang

The primary goal of graph knowledge distillation (GKD) is to transfer knowledge from a complex graph neural network (GNN) teacher to a smaller, yet more efficient GNN or multi-layer perceptron student. Although existing methods address network scalability, they rely on a frozen teacher that fails to explain how to derive results, thus limiting performance and hindering the improvement of a student. Therefore, we propose a novel GKD method, termed Dynamic Gradient Distillation (DGD), consisting of Generative Adversarial Imitation Learning (GAIL)-based Gradient Mapping and Two-Stage Gradient Fusion modules. The former builds the teacher’s learning process to understand knowledge by drawing on the principle of GAIL. The latter consists of attention fusion and weighted bias operations. Through the attentional fusion operation, it captures and fuses the responses of the teacher to change the gradient of the student at each layer. The fused gradients are then updated by combining them with the student’s backpropagated gradients using the weighted bias operation. DGD allows the student to inherit and extend the teacher’s learning process efficiently. Extensive experiments conducted with seven publicly available datasets show that DGD could significantly outperform some existing methods in node classification tasks. Our code and data are released at https://github.com/KangL-G/Dynamic-Gradient-Distillation.

图知识蒸馏（GKD）的主要目标是将知识从复杂图神经网络（GNN）教师转移到更小但更高效的GNN或多层感知器学生。虽然现有的方法解决了网络的可扩展性，但它们依赖于一个僵化的老师，无法解释如何得出结果，从而限制了学生的表现，阻碍了学生的进步。因此，我们提出了一种新的GKD方法，称为动态梯度蒸馏（DGD），由基于生成对抗模仿学习（GAIL）的梯度映射和两阶段梯度融合模块组成。前者借鉴了GAIL的原理，构建了教师理解知识的学习过程。后者包括注意融合和加权偏置操作。通过注意融合操作，捕捉并融合教师的反应，改变学生在每一层的梯度。然后使用加权偏置操作将融合的梯度与学生的反向传播梯度结合起来进行更新。DGD允许学生有效地继承和扩展老师的学习过程。在7个公开数据集上进行的大量实验表明，在节点分类任务中，DGD可以显著优于一些现有的方法。我们的代码和数据发布在https://github.com/KangL-G/Dynamic-Gradient-Distillation。

{"title":"A novel knowledge distillation method for graph neural networks with gradient mapping and fusion","authors":"Kang Liu , Shunzhi Yang , Chang-Dong Wang , Yunwen Chen , Zhenhua Huang","doi":"10.1016/j.inffus.2026.104163","DOIUrl":"10.1016/j.inffus.2026.104163","url":null,"abstract":"<div><div>The primary goal of graph knowledge distillation (GKD) is to transfer knowledge from a complex graph neural network (GNN) teacher to a smaller, yet more efficient GNN or multi-layer perceptron student. Although existing methods address network scalability, they rely on a frozen teacher that fails to explain how to derive results, thus limiting performance and hindering the improvement of a student. Therefore, we propose a novel GKD method, termed Dynamic Gradient Distillation (DGD), consisting of Generative Adversarial Imitation Learning (GAIL)-based Gradient Mapping and Two-Stage Gradient Fusion modules. The former builds the teacher’s learning process to understand knowledge by drawing on the principle of GAIL. The latter consists of attention fusion and weighted bias operations. Through the attentional fusion operation, it captures and fuses the responses of the teacher to change the gradient of the student at each layer. The fused gradients are then updated by combining them with the student’s backpropagated gradients using the weighted bias operation. DGD allows the student to inherit and extend the teacher’s learning process efficiently. Extensive experiments conducted with seven publicly available datasets show that DGD could significantly outperform some existing methods in node classification tasks. Our code and data are released at <span><span>https://github.com/KangL-G/Dynamic-Gradient-Distillation</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50367,"journal":{"name":"Information Fusion","volume":"131 ","pages":"Article 104163"},"PeriodicalIF":15.5,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146006526","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