检索结果-内蒙古大学图书馆

SSRN 2024年

作者： Zhang, Xin Wang, Bing Huang, Taifeng Yang, Shuo Tian, Xuedong Zhai, Junhai Yang, Ying College of Mathematics and Information Science Hebei University Wusi Road 180 Hebei Baoding071000 China Hebei Key Laboratory of Machine Learning and Computational Intelligence Hebei University Wusi Road 180 Hebei Baoding071000 China Hebei University Affiliated Hospital Hebei University Wusi Road 180 Hebei Baoding071000 China College of Electronic Information Engineering Hebei University Qiyi Road 2666 Hebei Baoding071000 China Hebei University China

Semi-supervised learning (SSL) has emerged as one of effective strategies to leverage useful information from unlabeled datasets to improve deep model performance, and it is widely utilized in medical image segmentation to alleviate the burden of expensive pixel-level labeling. Most existing SSL-based medical image segmentation methods utilize all unlabeled data equally to update the models through unsupervised loss. However, not all unlabeled data is applicable to the same training process of the model due to varying data quality and importance. In this paper, we propose a new dual-decoder data decoupling training-based semi-supervised medical image segmentation network (DD-Net), allowing the model to focus on challenging regions. DD-Net decouples the prediction of unlabeled data into data with different functions based on the confidence matching degree from two student decoders, and adopts different optimization strategies for different functional data. Specifically, for high-reliability part, we employ the cross pseudo supervision learning to improve the reliability of the model prediction. For divergent prediction, we propose mutual matching learning to guide the model to learn richer information from high uncertainty by assigning high entropy to the prediction data. For the low confidence prediction, we present the reinforcement consistency learning to enhance the model context representations to effectively capture the important details, such as edges and contours, from unlabeled data. Experiments on four medical challenging image datasets show that our method outperforms the previous state-of-the-art methods in the field. The code is available at https://***/TaifengHuang/DD-Net. © 2024, The Authors. All rights reserved.

关键词： Forecasting

来源：评论

学校读者我要写书评

暂无评论

Ma-Transunet: U-Shaped Transformer with Multi-Scale Cnn-Based Auxiliary Network for Medical Image Segmentation

SSRN

引用

SSRN 2024年

作者： Wang, Bing Zhao, Zutong Wei, Zhihong Zhai, Junhai Tian, Xuedong Zhang, Xin Yang, Ying College of Mathematics and Information Science Hebei University Wusi Road 180 Hebei Baoding071000 China Hebei Key Laboratory of Machine Learning and Computational Intelligence Hebei University Wusi Road 180 Hebei Baoding071000 China Hebei University Affiliated Hospital Hebei University Wusi Road 180 Hebei Baoding071000 China College of Electronic Information Engineering Hebei University Qiyi Road 2666 Hebei Baoding071000 China Hebei University China

Recently, pure Transformer-based U-shape networks (PTUs) have achieved significant success in medical image segmentation tasks typically dominated by Convolutional Neural Networks (CNNs), due to their outstanding ability to capture global long-range dependencies at multiple layers. However, they still suffer from limitations in local detail representation such as textures and edges. While some CNN-Transformer complementary approaches leverage the merits of both to obtain higher performance, they either fail to fully exploit the benefits brought by PTU models or adopt too complex feature fusion manners that can reduce network flexibility and limit their practice in clinic. To address these issues, we propose a novel CNN-Transformer combined approach (MA-TransUnet), which is a main-auxiliary separable framework. U-shaped Transformer main network (UTM) captures multi-level global context correlations, and multi-scale CNN-based auxiliary network (MCA) remedies local detail extraction and ensures semantic feature consistency. The separable design style of MA-TransUnet allows it to flexibly adapt to a variety of existing Transformer blocks and structures, facilitating to efficiently build the combined networks based on Transformer and CNN to improve the performance of the whole segmentation models. Besides, we propose a novel multi-scale Transformer block based on efficient attention mechanism to further enhance multi-scale feature extraction for MA-TransUnet network. Extensive experiments on synapse multi-organ and ISIC 2018 skin lesion segmentation datasets demonstrate that the MA-TransUnet surpasses the majority of current methods based solely on CNNs, Transformers, or hybrid CNN-Transformers with good efficiency and generality. Code is available on https://***/Zutong-Zhao/***. © 2024, The Authors. All rights reserved.

关键词： Extraction

来源：评论

学校读者我要写书评

暂无评论

A Deep Model Towards Accurate Boundary Location and Strong Generalization for Medical Image Segmentation

SSRN

引用

SSRN 2023年

作者： Wang, Bing Geng, Peipei Li, Tianxu Yang, Ying Tian, Xuedong Zhang, Guochun Zhang, Xin College of Mathematics and Information Science Hebei University Hebei Baoding071000 China Hebei Key Laboratory of Machine Learning and Computational Intelligence Hebei University Hebei Baoding071000 China Hebei University Affiliated Hospital Hebei Baoding071000 China College of Cyber Security and Computer Hebei University Hebei Baoding071000 China College of Electronic Information Engineering Hebei University Hebei Baoding071000 China

Accurate medical image segmentation plays a crucial role in computer-assisted diagnosis and monitoring. However, due to the complexity of medical images and the limitations of image acquisition, most of the current segmentation models still face the challenges of imprecise boundary positioning and poor generalization ability, which greatly affect their application in clinical practices. To improve the accuracy of object boundary location, a Local-global module is proposed, which uses attention mechanism to model intra-sample and inter-sample long-range dependencies from local and global perspectives respectively. Meanwhile, we exploit a convolutional neural network with a generalization enhancement module which is able to adaptively adjust weights of the classifier during test to improve the generalization ability of the model. Furthermore, we design a multi-granularity loss function which can guide the network to supervise regions and boundary details from pixel-level, patch-level, and map-level, respectively. To verify the effectiveness of our proposed method, we conduct extensive experiments on five medical image datasets of different modalities, including complex lung tumors, cell nuclei, brain tumors, skin lesions, and colon polyps. Our method achieves better performance on the Dice coefficient and Hausdorff distance. We further verify the generalization ability of the model on the colon polyp dataset (including unseen sub-datasets in the testing set). © 2023, The Authors. All rights reserved.

关键词： Image segmentation

来源：评论

学校读者我要写书评

暂无评论

A Novel ECG Signal Classification Algorithm Based on Common and Specific Components Separation 1

引用

2nd International Conference on Pattern Recognition and Artificial Intelligence, ICPRAI 2020

作者： Huang, Jianfeng Huang, Chao Yang, Lihua Zhang, Qian School of Financial Mathematics and Statistics Guangdong University of Finance Guangzhou China College of Mathematics and Statistics Shenzhen Key Laboratory of Advanced Machine Learning and Applications Shenzhen University Shenzhen China Guangdong Province Key Laboratory of Computational Science School of Mathematics Sun Yat-sen University Guangzhou China

ISBN: (数字)9783030598303

ISBN: (纸本)9783030598297

Electrocardiography (ECG) signal classification is a challenging task since the characteristics of ECG signals vary significantly for different patients. In this paper, we propose a new method for ECG signal classification based on the separation of common and specific components of a signal. The common components are obtained via Canonical Correlation Analysis (CCA). After removing the common components from the signal, we map the specific components to a lower dimensional feature space for classification. We first establish a basic model in the binary classification setting and then extend it to a more general version. Numerical experiments results on the MIT-BIH Arrhythmia Database are presented and discussed. © 2020, Springer Nature Switzerland AG.

关键词： Electrocardiography

来源：评论

学校读者我要写书评

暂无评论

Global well-posedness and decay estimates for the 2D chemotaxis model with the supercritical case

引用

Journal of Mathematical Analysis and Applications 2025年第1期552卷

作者： Peiyi Liu Qian Zhang Chinese Academy of Inspection & Quarantine Greater Bayarea Zhongshan 528000 PR China Hebei Key Laboratory of Machine Learning and Computational Intelligence College of Mathematics and Information Science Hebei University Baoding 071002 PR China

In this paper, we investigate the Cauchy problem for the 2D chemotaxis model. By considering an external force term of the form − ρ q for 1 ≤ q < 2 , we establish the existence of a global classical solution and ... 详细信息

关键词： Chemotaxis Decay estimates Global solutions

来源：评论

学校读者我要写书评

暂无评论

Dual-consistency guidance semi-supervised medical image segmentation with low-level detail feature augmentation

引用

Computers in Biology and Medicine 2024年 181卷 109046-109046页

作者： Wang, Bing Ju, Mengyi Zhang, Xin Yang, Ying Tian, Xuedong College of Mathematics and Information Science Hebei University Wusi Road 180 Hebei Baoding071000 China Hebei Key Laboratory of Machine Learning and Computational Intelligence Hebei University Wusi Road 180 Hebei Baoding071000 China College of Electronic Information Engineering Hebei University Qiyi Road 2666 Hebei Baoding071000 China Hebei University Affiliated Hospital Hebei University Wusi Road 180 Hebei Baoding071000 China College of Cyber Security and Computer Hebei University Wusi Road 180 Hebei Baoding071000 China

In deep-learning-based medical image segmentation tasks, semi-supervised learning can greatly reduce the dependence of the model on labeled data. However, existing semi-supervised medical image segmentation methods face the challenges of object boundary ambiguity and a small amount of available data, which limit the application of segmentation models in clinical practice. To solve these problems, we propose a novel semi-supervised medical image segmentation network based on dual-consistency guidance, which can extract reliable semantic information from unlabeled data over a large spatial and dimensional range in a simple and effective manner. This serves to improve the contribution of unlabeled data to the model accuracy. Specifically, we construct a split weak and strong consistency constraint strategy to capture data-level and feature-level consistencies from unlabeled data to improve the learning efficiency of the model. Furthermore, we design a simple multi-scale low-level detail feature enhancement module to improve the extraction of low-level detail contextual information, which is crucial to accurately locate object contours and avoid omitting small objects in semi-supervised medical image dense prediction tasks. Quantitative and qualitative evaluations on six challenging datasets demonstrate that our model outperforms other semi-supervised segmentation models in terms of segmentation accuracy and presents advantages in terms of generalizability. Code is available at https://***/0Jmyy0/SSMIS-DC. © 2024 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

Primal Characterizations of Error Bounds for Composite-convex Inequalities

arXiv

引用

arXiv 2022年

作者： Wei, Zhou Théra, Michel Yao, Jen-Chih Hebei Key Laboratory of Machine Learning and Computational Intelligence College of Mathematics and Information Science Hebei University Baoding071002 China Xlim UMR-CNRS 7252 Université de Limoges Limoges France Federation University Australia Ballarat Australia Department of Applied Mathematics National Sun Yat-sen University Kaohsiung Taiwan

This paper is devoted to primal conditions of error bounds for a general function. In terms of Bouligand tangent cones, lower Hadamard directional derivatives and the Hausdorff-Pompeiu excess of subsets, we provide several necessary and/or sufficient conditions of error bounds with mild assumptions. Then we use these primal results to characterize error bounds for composite-convex functions (i.e. the composition of a convex function with a continuously differentiable mapping). It is proved that the primal characterization of error bounds can be established via Bouligand tangent cones, directional derivatives and the Hausdorff-Pompeiu excess if the mapping is metrically regular at the given point. The accurate estimate on the error bound modulus is also obtained. © 2022, CC0.

关键词： Mapping

来源：评论

学校读者我要写书评

暂无评论

Ct-Gan: A Conditional Generative Adversarial Network of Transformer Architecture for Text-to-Image

SSRN

引用

SSRN 2022年

作者： Wang, Bing Zhang, Xin Jiao, Wentao Tian, Xuedong College of Electronic Information Engineering Hebei University Hebei Baoding071000 China Hebei Key Laboratory of Machine Learning and Computational Intelligence Hebei University Hebei Baoding071000 China College of Mathematics and Information Science Hebei University Hebei Baoding071000 China College of Cyber Security and Computer Hebei University Hebei Baoding071000 China

How to generate an image from a text description is an imaginative and challenging task. This study proposes a conditional generative adversarial network (GAN) of transformer architecture for text-to-image tasks called CT-GAN by employing the GAN generator based on transformer architecture. We also propose a filtering module suitable for non-end-to-end multi-stage models. This module can screen out the good images generated in the previous stage and allows only the good images to participate in the generation of the later stage. This method significantly improves the quality of the generated images. Furthermore, we designed a generator and discriminator based on symmetry. In the generator, we propose a shift self-attention technology to establish information communication between grids, reduce boundary loss, and improve image quality. We established two modes of local and global discriminations based on the grid, which can balance the performance of the generator and discriminator, improve the training stability, and accelerate the model convergence. We conducted several experiments on the widely used conditional datasets (CUB and COCO) and unconditional datasets (CelebA and LSUN church). The experimental results show that the proposed CT-GAN is superior to the most advanced convolution model in generating diversity and semantic consistency. © 2022, The Authors. All rights reserved.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

A hybrid monotone decision tree model for interval-valued attributes

引用

Advances in computational Intelligence 2021年第1期2卷 1-11页

作者： Chen, Jiankai Li, Zhongyan Wang, Xin Zhai, Junhai School of Control and Computer Engineering North China Electric Power University Beijing China Hebei Key Laboratory of Machine Learning and Computational Intelligence College of Mathematics and Information Science Hebei University Baoding China School of Mathematics and Physics North China Electric Power University Beijing China

The existing monotonic decision tree algorithms are based on a linearly ordered constraint that certain attributes are monotonously consistent with the decision, which could be called monotonic attributes, whereas others, called non-monotonic attributes. In practice, monotonic and non-monotonic attributes coexist in most classification tasks, and some attribute values are even evaluated as interval numbers. In this paper, we proposed a fuzzy rank-inconsistent rate based on probability degree to judge the monotonicity of interval numbers. Furthermore, we devised a hybrid model composed of monotonic and non-monotonic attributes to construct a mixed monotone decision tree for interval-valued data. Experiments on artificial and real-world data sets show that the proposed hybrid model is effective.

关键词：

来源：评论

学校读者我要写书评

暂无评论

High-speed and low-power molecular dynamics processing unit(MDPU)with ab initio accuracy

引用

npj computational Materials 2024年第1期10卷 559-568页

作者： Pinghui Mo Yujia Zhang Zhuoying Zhao Hanhan Sun Junhua Li Dawei Guan Xi Ding Xin Zhang Bo Chen Mengchao Shi Duo Zhang Denghui Lu Yinan Wang Jianxing Huang Fei Liu Xinyu Li Mohan Chen Jun Cheng Bin Liang Weinan E Jiayu Dai Linfeng Zhang Han Wang Jie Liu College of Integrated Circuits Hunan UniversityChangsha410082P.R.China College of Electrical and Information Engineering Hunan UniversityChangsha410082P.R.China College of Computer National University of Defense TechnologyChangsha410073P.R.China Department of Physics National University of Defense TechnologyChangsha410073P.R.China DP Technology 100080 BeijingP.R.China AI for Science Institute 100080 BeijingP.R.China Academy for Advanced Interdisciplinary Studies Peking University100871 BeijingP.R.China HEDPS CAPTCollege of EngineeringPeking University100871 BeijingP.R.China School of Mathematical Science Peking University100871 BeijingP.R.China State Key Laboratory of Physical Chemistry of Solid Surfaces iChEMCollege of Chemistry and Chemical EngineeringXiamen UniversityXiamen361005P.R.China School of Integrated Circuits Peking University100871 BeijingP.R.China Center for Machine Learning Research Peking University100871 BeijingP.R.China Institute of Applied Physics and Computational Mathematics 100088 BeijingP.R.China Greater Bay Area Institute for Innovation Hunan UniversityGuangzhou511300P.R.China Department of Electrical and Computer Engineering University of WashingtonSeattleWA98195USA

Molecular dynamics(MD)is an indispensable atomistic-scale computational tool widely-used in various *** the past decades,nearly all ab initio MD and machine-learning MD have been based on the general-purpose central/graphics processing units(CPU/GPU),which are well-known to suffer from their intrinsic“memory wall”and“power wall”***,nowadays MD calculations with ab initio accuracy are extremely time-consuming and power-consuming,imposing serious restrictions on the MD simulation size and *** solve this problem,here we propose a special-purpose MD processing unit(MDPU),which could reduce MD time and power consumption by about 103 times(109 times)compared to state-of-the-art machine-learningMD(ab initio MD)based on CPU/GPU,while keeping ab initio *** significantly-enhanced performance,the proposed MDPU may pave a way for the accurate atomistic-scale analysis of large-size and/or longduration problems which were impossible/impractical to compute before.

关键词： consuming power MDP

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：