检索结果-内蒙古大学图书馆

Complementary Learning Subnetworks Towards Parameter-Efficient Class-Incremental Learning

IEEE Transactions on Knowledge and Data Engineering 2025年第6期37卷 3240-3252页

作者： Li, Depeng Zeng, Zhigang Dai, Wei Suganthan, Ponnuthurai Nagaratnam Huazhong University of Science and Technology School of Artificial Intelligence and Automation Wuhan430074 China Key Laboratory of Image Processing and Intelligent Control of Education Ministry of China Wuhan430074 China China University of Mining and Technology School of Information and Control Engineering Xuzhou221116 China Qatar University Kindi Center for Computing Research College of Engineering Doha2713 Qatar

In the scenario of class-incremental learning (CIL), deep neural networks have to adapt their model parameters to non-stationary data distributions, e.g., the emergence of new classes over time. To mitigate the catastrophic forgetting phenomenon, typical CIL methods either cumulatively store exemplars of old classes for retraining model parameters from scratch or progressively expand model size as new classes arrive, which, however, compromises their practical value due to little attention paid to parameter efficiency. In this paper, we contribute a novel solution, effective control of the parameters of a well-trained model, by the synergy between two complementary learning subnetworks. Specifically, we integrate one plastic feature extractor and one analytical feed-forward classifier into a unified framework amenable to streaming data. In each CIL session, it achieves non-overwritten parameter updates in a cost-effective manner, neither revisiting old task data nor extending previously learned networks;Instead, it accommodates new tasks by attaching a tiny set of declarative parameters to its backbone, in which only one matrix per task or one vector per class is kept for knowledge retention. Experimental results on a variety of task sequences demonstrate that our method achieves competitive results against state-of-the-art CIL approaches, especially in accuracy gain, knowledge transfer, training efficiency, and task-order robustness. Furthermore, a graceful forgetting implementation on previously learned trivial tasks is empirically investigated to make its non-growing backbone (i.e., a model with limited network capacity) suffice to train on more incoming tasks. © 2025 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Multi-Level Speaker Representation for Target Speaker Extraction

Multi-Level Speaker Representation for Target Speaker Extrac...

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025

作者： Zhang, Ke Li, Junjie Wang, Shuai Wei, Yangjie Wang, Yi Wang, Yannan Li, Haizhou Key Laboratory of Intelligent Computing in Medical Image Northeastern University China SDS SRIBD The Chinese University of Hong Kong Shenzhen China The Hong Kong Polytechnic University Hong Kong Tencent Ethereal Audio Lab Tencent Shenzhen China University of Bremen Germany Department of Electrical and Computer Engineering National University of Singapore Singapore

ISBN: (纸本)9798350368741

Target speaker extraction (TSE) relies on a reference cue of the target to extract the target speech from a speech mixture. While a speaker embedding is commonly used as the reference cue, such embedding pre-trained with a large number of speakers may suffer from confusion of speaker identity. In this work, we propose a multi-level speaker representation approach, from raw features to neural embeddings, to serve as the speaker reference cue. We generate a spectral-level representation from the enrollment magnitude spectrogram as a raw, low-level feature, which significantly improves the model's generalization capability. Additionally, we propose a contextual embedding feature based on cross-attention mechanisms that integrate frame-level embeddings from a pre-trained speaker encoder. By incorporating speaker features across multiple levels, we significantly enhance the performance of the TSE model. Our approach achieves a 2.74 dB improvement and a 4.94% increase in extraction accuracy on Libri2mix test set over the baseline. © 2025 IEEE.

关键词： Cocktail party problem selective auditory attention speaker confusion speaker feature target speaker extraction

来源：评论

学校读者我要写书评

暂无评论

REMOTE: Real-time Ego-motion Tracking for Various Endoscopes via Multimodal Visual Feature Learning

arXiv

引用

arXiv 2025年

作者： Shao, Liangjing Chen, Benshuang Zhao, Shuting Chen, Xinrong Academy for Engineering & Technology Fudan University China Shanghai Key Laboratory of Medical Image Computing and Computer-Assisted Intervention Fudan University China

Real-time ego-motion tracking for endoscope is a significant task for efficient navigation and robotic automation of endoscopy. In this paper, a novel framework is proposed to perform real-time ego-motion tracking for endoscope. Firstly, a multi-modal visual feature learning network is proposed to perform relative pose prediction, in which the motion feature from the optical flow, the scene features and the joint feature from two adjacent observations are all extracted for prediction. Due to more correlation information in the channel dimension of the concatenated image, a novel feature extractor is designed based on an attention mechanism to integrate multi-dimensional information from the concatenation of two continuous frames. To extract more complete feature representation from the fused features, a novel pose decoder is proposed to predict the pose transformation from the concatenated feature map at the end of the framework. At last, the absolute pose of endoscope is calculated based on relative poses. The experiment is conducted on three datasets of various endoscopic scenes and the results demonstrate that the proposed method outperforms state-of-the-art methods. Besides, the inference speed of the proposed method is over 30 frames per second, which meets the real-time requirement. The project page is here: *** © 2025, CC BY.

关键词： Endoscopy

来源：评论

学校读者我要写书评

暂无评论

A Hybrid-Feature-Based Autoencoder Model for Predicting Traffic Flow to Accelerate Ambulance medical Response Time in Internet of Vehicles

引用

IEEE Internet of Things Journal 2025年第9期12卷 11397-11407页

作者： Li, Yongmeng Tian, Yu Liu, Wenjian Gao, Longxiang Tian, Hui Xing, Lumin The First Affiliated Hospital of Shandong First Medical University Shandong Provincial Qianfoshan Hospital Shandong Key Laboratory of Digital Diagnosis and Treatment of Thoracic Oncology Shandong Engineering Research Center of Intelligent Surgery Shandong Jinan250014 China Qilu Hospital of Shandong University Department of Thoracic Surgery Shandong Jinan250012 China City University of Macau Faculty of Data Science China Qilu University of Technology Shandong Academy of Sciences Key Laboratory of Computing Power Network and Information Security Ministry of Education Shandong Computer Science Center Jinan250014 China Shandong Fundamental Research Center for Computer Science Shandong Provincial Key Laboratory of Computer Networks Jinan250000 China The First Affiliated Hospital of Shandong First Medical University Shandong Provincial Qianfoshan Hospital Shandong Key Laboratory of Digital Diagnosis and Treatment of Thoracic Oncology Shandong Jinan250014 China

Traffic congestion during ambulance travel can delay medical response times. With the growing availability of traffic data, computational methods for traffic flow prediction are attracting significant attention. However, current computational prediction models have the following shortcomings. Fully connected networks require extensive feature engineering yet struggle to capture local traffic flow features. Their high parameter count increases training time and overfitting risk, while varying traffic conditions hinder model generalization. Therefore, in this work, we propose a hybrid-feature-based autoencoder (HFAE) model to predict traffic flow and accelerate ambulance medical response time in Internet of Vehicles. Specifically, the HFAE model simultaneously considers local features from the convolutional neural network and global features from the fully connected neural network. Additionally, the HFAE model uses an autoencoder to reconstruct the original input, capturing the intrinsic structure of the traffic flow data. Specifically, the root mean-squared error and mean absolute error scores of the HFAE model are better than those of the three mainstream models. © 2014 IEEE.

关键词： Ambulances

来源：评论

学校读者我要写书评

暂无评论

Online Queue-Aware Service Migration and Resource Allocation in Mobile Edge computing

引用

IEEE Transactions on Vehicular Technology 2025年

作者： Du, An Jia, Jie Chen, Jian Wang, Xingwei Huang, Ming Northeastern University School of Computer Science and Engineering Engineering Research Center of Security Technology of Complex Network System Key Laboratory of Intelligent Computing in Medical Image Ministry of Education Shenyang110819 China Northeastern University School of Computer Science and Engineering Shenyang110819 China

Mobile edge computing (MEC) integrated with Network Functions Virtualization (NFV) helps run a wide range of services implemented by Virtual Network Functions (VNFs) deployed at MEC networks. This emerging paradigm offers flexible edge resource management for Internet Service Providers (ISPs) and improves service satisfaction of diverse applications. However, a critical challenge lies in processing extremely unpredictable and bursty traffic flow, especially in event-trigger sensing tasks. In this paper, we study queue-aware service migration and resource allocation in an MEC network, where different devices with different arrival patterns request different network services. To this end, we first formulate a novel long-term operation cost minimization problem subject to the stability of multiple queues by adopting a simultaneous migration scheme and determining resource allocation dynamically. To address the problem efficiently, we propose an online control framework to make decisions without any prior system information. We first adopt the Lyapunov optimization technique to decompose the long-term optimization problem into multi-per-slot subproblems. Then, we adopt a decoupled optimization algorithm based on the coordinate descent method and augmented Lagrangian method to solve the offline mixed integer non-linear programming (MINLP) subproblems. Extensive results reveal that the proposed online optimization framework can efficiently balance queue stability and cost consumption. © 1967-2012 IEEE.

关键词： Mobile edge computing

来源：评论

学校读者我要写书评

暂无评论

An Adaptive Federated Domain Generalization Framework for Consumer Electronics Manufacturing Equipment Cross-Factory Fault Detection

引用

IEEE Transactions on Consumer Electronics 2025年

作者： Li, Haodong Wang, Xingwei Li, Ying Yi, Bo Cao, Peng Huang, Min Li, Keqin Northeastern University College of Computer Science and Engineering Shenyang China Northeastern University Key Laboratory of Intelligent Computing in Medical Image Shenyang China Northeastern University College of Information Science and Engineering Shenyang China Hunan University College of Computer Science and Electronic Engineering Changsha China

Reliability and operational efficiency of equipment are crucial in the manufacturing of consumer electronics. Existing fault detection methods often face limitations such as dataset dependence, poor scenario generalization, and data privacy issues when addressing the complex and diverse operating conditions in product manufacturing. To address these issues, this paper proposes a cross-factory fault detection framework for consumer electronics production equipment based on adaptive federated domain generalization. This framework reconsiders the limitations of Sharpness-Aware Minimization(SAM) and, by jointly considering local personalization and global generalization objectives, designs an adaptive weighting scheme to balance the trade-off between loss minimization and sharpness during optimization, thereby improving the model's robustness and accuracy under various working conditions. Then, A parameter momentum aggregation scheme is proposed on the server side to incorporate historical gradient information, reducing client drift impact and improving model convergence and stability. Finally, extensive scenario experiments were conducted on two public datasets. The results indicate that the proposed framework achieves an average improvement of 22.5% in fault detection accuracy over the baseline model across varying operating conditions and data distribution scenarios, demonstrating its effectiveness in addressing the challenges of complex condition variations and data privacy in consumer electronics manufacturing. © 1975-2011 IEEE.

关键词： Smart manufacturing

来源：评论

学校读者我要写书评

暂无评论

Uncertainty Quantification and Quality Control for Heatmap-based Landmark Detection Models

引用

IEEE Transactions on medical Imaging 2025年

作者： Feng, Yong Yang, Jinzhu Tang, Lingzhi Sun, Song Wang, Yonghuai Northeastern University Key Laboratory of Intelligent Computing in Medical Image Ministry of Education Shenyang China Northeastern University School of Computer Science and Engineering Shenyang China National Frontiers Science Center for Industrial Intelligence and Systems Optimization Shenyang China The First Hospital of China Medical University Department of Cardiovascular Ultrasound Shenyang China

Uncertainty quantification is a vital aspect of explainable artificial intelligence that fosters clinician trust in medical applications and facilitates timely interventions, leading to safer and more reliable outcomes. Although deep learning models have reached clinically acceptable accuracy in anatomical landmark detection, their predictions remain susceptible to contextual noise due to the small size of the target structures, making uncertainty quantification more challenging than in classification and segmentation tasks. This paper presents an end-to-end uncertainty quantification method tailored for heatmapbased anatomical landmark detection models, designed to improve both interpretability and controllability in clinical applications. Leveraging Dempster-Shafer Theory and Subjective Logic Theory, we implement probability assignment and uncertainty quantification through a single forward pass to ensure computational efficiency. We introduce an evidence map that captures the strength of landmark evidence, alongside an uncertainty map that calibrates predicted probabilities within the Subjective Logic framework. The interaction between these two components, facilitated by a cross-attention mechanism, further improves landmark detection accuracy and enhances the effectiveness of uncertainty quantification. Experimental results demonstrate that the proposed method maintains detection accuracy, even in noisy environments, while outperforming state-of-the-art methods in terms of uncertainty quantification and quality control. Furthermore, the model effectively identifies out-of-distribution data solely through calibrated probabilities when encountering inconsistencies in multi-center data and novel data, underscoring its potential for clinical applications. © 1982-2012 IEEE.

关键词： Model predictive control

来源：评论

学校读者我要写书评

暂无评论

A semi-supervised multi-task assisted method for ultrasound medical image segmentation

引用

Neurocomputing 2025年 639卷

作者： Li, Honghe Yang, Jinzhu Qu, Mingjun Feng, Yong Key Laboratory of Intelligent Computing in Medical Image Ministry of Education Northeastern University Liaoning Shenyang110819 China School of Computer Science and Engineering Northeastern University Liaoning Shenyang110819 China National Frontiers Science Center for Industrial Intelligence and Systems Optimization Liaoning Shenyang110819 China

The accurate segmentation of the left ventricle in echocardiography is critical for assessing cardiac function, but challenges such as blurred boundaries, high morphological variability, and limited annotated data hinder the performance of traditional methods. Recent advances in semi-supervised learning show promise in leveraging unlabeled data, yet existing techniques often rely on network or data-level perturbation, which may not fully exploit spatial and positional information essential for precise segmentation. To address these challenges, we propose a novel multi-task assisted semi-supervised segmentation framework. Our method combines segmentation, landmark detection, and image reconstruction into a unified model with a shared encoder and dual decoders. A feature cross-fusion module based on a cross-attention mechanism integrates features across tasks to enhance spatial and positional awareness. We have also introduced a contrastive learning mechanism to refine the segmentation boundaries, especially in areas where the edges are blurred. In addition, we utilize multi-scale supervision to better adapt the model to targets of different scales. The framework employs an exponential moving average student–teacher model to effectively utilize unlabeled data for training. Experiments on CAMUS and EchoNet-Dynamic datasets demonstrate that the proposed method achieves state-of-the-art performance, delivering near fully-supervised results with only 10%–20% labeled data. This highlights its potential for high-quality segmentation in low-annotation scenarios, outperforming existing semi-supervised learning methods. © 2025

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

MCHNet: An Efficient Cross Attention-Guided Hierarchical Multiscale Network for Segmentation of Organs at Risk in CT images

引用

IEEE Transactions on Radiation and Plasma medical Sciences 2025年第5期9卷 598-612页

作者： Guo, Huimin Gu, Yin Du, Wu Chen, Boyang Cui, Ming Zhang, Teng Sun, Deyu Qian, Wei Ma, He Northeastern University College of Medicine and Biological Information Engineering Shenyang110169 China Cancer Hospital of Dalian University of Technology Department of Radiation Oncology Gastrointestinal and Urinary and Musculoskeletal Cancer Shenyang110042 China Northeastern University College of Medicine and Biological Information Engineering Key Laboratory of Intelligent Computing in Medical Image Ministry of Education Shenyang110169 China

In radiotherapy, precisely contoured organs at risk (OARs) near the target areas are essential for effective treatment planning. Manual delineation of OARs is labor-intensive and varies among experts. Deep learning has improved accuracy and consistency, but current methods use complex architectures with many parameters, risking overfitting on small medical image datasets. Addressing this, this article introduces MCHNet, a cross attention (CA)-guided hierarchical multiscale segmentation network based on CT images. MCHNet adopts a typical U-shaped structure, with MobileVit as its backbone. An efficient CA-guided block is utilized to enhance feature extraction while minimizing model parameters. Additionally, a novel skip-connection strategy is proposed to preserve critical medical image information during multiple down-sampling operations and bridge the gap between deep and shallow features. We conducted extensive experiments on three public datasets, i.e., multiatlas labeling beyond the cranial vault dataset, Segmentation of thoracic OARs dataset, and multimodality abdominal multiorgan segmentation challenge 2022 dataset. Experiment results demonstrate that the proposed method outperforms other related Transformer-based or hybrid models in terms of computational complexity, quantitative and qualitative results. We believe that the proposed method can offer an optimal blend of precision and efficiency that advances the capabilities of OARs segmentation in radiotherapy. © 2017 IEEE.

关键词： image segmentation

来源：评论

学校读者我要写书评

暂无评论

NaMA-Mamba: Foundation model for generalizable nasal disease detection using masked autoencoder with Mamba on endoscopic images

引用

Computerized medical Imaging and Graphics 2025年 122卷 102524-102524页

作者： Wang, Wensheng Jin, Zewen Liu, Xueli Chen, Xinrong Academy for Engineering and Technology Fudan University Shanghai200433 China Shanghai Key Laboratory of Medical Image Computing and Computer Assisted Intervention Shanghai200032 China Eye & ENT Hospital of Fudan University Shanghai200031 China

Artificial intelligence (AI) has shown great promise in analyzing nasal endoscopic images for disease detection. However, current AI systems require extensive expert-labeled data for each specific medical condition, limiting their applications. In this work, the challenge is addressed through two key innovations, the creation of the first large-scale pre-training dataset of nasal endoscopic images, and the development of a novel self-learning AI system specifically designed for nasal endoscopy, named NaMA-Mamba. In the proposed NaMA-Mamba model, two key technologies are utilized, which are the nasal endoscopic state space model (NE-SSM) for analyzing sequences of images and an enhanced learning mechanism (CoMAE) for capturing fine details in nasal tissues. These innovations enable the system to learn effectively from unlabeled images while maintaining high accuracy across different diagnostic tasks. In extensive testing, NaMA-Mamba achieved remarkable results using minimal labeled data, matching the performance of traditional systems that require full expert labeling while needing only 1% of the labeled data for tasks such as detecting nasal polyps and identifying nasopharyngeal cancer. These results demonstrate the potential of NaMA-Mamba to significantly improve the efficiency and accessibility of AI-assisted nasal disease diagnosis in clinical practice. © 2025 Elsevier Ltd

关键词： Self-supervised learning

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：