检索结果-内蒙古大学图书馆

Deep Learning Model for Analyzing EEG Signal Analysis

IEEE Access 2025年 13卷 91034-91045页

作者： Gupta, Varun Kumar, Vivek Prince Singh, Saurabh Lee, Young-Seok Ra, In-Ho Department of Electronics and Communication National Institute of Technology Sikkim India National Institute of Technology Department of Electrical and Electronics Engineering Sikkim India Dongguk University Department of Industrial & Systems Engineering Seoul04620 Korea Republic of Woosong University Department of AI and Bigdata Daejeon34606 Korea Republic of Kunsan National University School of Computer & Software Gunsan Korea Republic of Kunsan National University School of Software Gunsan54150 Korea Republic of

To analyze the physiological information within the acquired EEG signal is very cumbersome due to the possibility of several factors, viz. noise and artifacts, complexity of brain dynamics, and inter-subject variability. To address these issues, this paper compares a U-shaped encoder-decoder network (UNET) and Bat-based UNET signal analysis (BUSA) techniques to classify depression rates in the Electroencephalogram (EEG) datasets. The main objective of including these two techniques is to reveal their effectiveness. It comprises pre-processing, feature extraction, feature selection, and classification stages. The framework excels at noise reduction during pre-processing, enhancing dataset integrity. Feature extraction leverages band power and correlation dimension to extract crucial features. Furthermore, feature selection optimizes classification accuracy by refining the fitness function of bats in the classification layer. The performance of UNET and BUSA are compared based on the following performance evaluating parameters viz. accuracy (Acc), Area Under the Curve (AUC), precision (P), and recall (R) (or sensitivity (Se)). The results indicated that the BUSA technique outperforms the UNET technique. © 2013 IEEE.

关键词： Electroencephalography

来源：评论

学校读者我要写书评

暂无评论

Joint high-resolution feature learning and vessel-shape aware convolutions for efficient vessel segmentation

引用

Computers in Biology and Medicine 2025年 191卷 109982-109982页

作者： Zhang, Xiang Zhu, Qiang Hu, Tao Guo, Song Bian, Genqing Dong, Wei Hong, Rao Lin, Xia Ling Wu, Peng Zhou, Meili Yan, Qingsen Mohi-ud-din, Ghulam Ai, Chen Li, Zhou College of Information and Control Engineering Xi'an University of Architecture and Technology Xi'an China Shaanxi Provincial Key Lab of Bigdata of Energy and Intelligence Processing School of Physics and Electronic Information Yanan University Yanan China School of Computer Science and Engineering Northwestern Polytechnical University Xi'an China School of Software Nanchang University Nanchang China School of Cyber Security Zhejiang University Hangzhou China Department of Basic Education and Research Jiangxi Police College Nanchang China Northwestern Polytechnical University China

Clear imagery of retinal vessels is one of the critical shreds of evidence in specific disease diagnosis and evaluation, including sophisticated hierarchical topology and plentiful-and-intensive capillaries. In this work, we propose a new topology- and shape-aware model named Multi-branch Vessel-shaped Convolution Network (MVCN) to adaptively learn high-resolution representations from retinal vessel imagery and thereby capture high-quality topology and shape information thereon. Two steps are involved in our pipeline. The former step is proposed as Multiple High-resolution Ensemble Module (MHEM) to enhance high-resolution characteristics of retinal vessel imagery via fusing scale-invariant hierarchical topology thereof. The latter is a novel vessel-shaped convolution that captures the retinal vessel topology to emerge from unrelated fundus structures. Moreover, our MVCN of separating such topology from the fundus is a dynamical multiple sub-label generation via using epistemic uncertainty, instead of manually separating raw labels to distinguish definitive and uncertain vessels. Compared to other existing methods, our method achieves the most advanced AUC values of 98.31%, 98.80%, 98.83%, and 98.65%, and the most advanced ACC of 95.83%, 96.82%, 97.09%,and 96.66% in DRIVE, CHASE_DB1, STARE, and HRF datasets. We also employ correctness, completeness, and quality metrics to evaluate skeletal similarity. Our method's evaluation metrics have doubled compared to previous methods, thereby demonstrating the effectiveness thereof. © 2025 Elsevier Ltd

关键词： Ophthalmology

来源：评论

学校读者我要写书评

暂无评论

Disentangling Long- and Short-term Preferences Based on Latent Commonalities Enhancement for Next Point-of-Interest Recommendation

引用

Expert Systems with Applications 2025年 284卷

作者： Zhang, Qing Huang, Xiao Zhang, Qi Zhou, Wei Wen, Junhao School of Bigdata and Software Engineering Chongqing University Daxuecheng South Rd. 55 Chongqing400044 China Key Laboratory of Dependable Service Computing in Cyber Physical Society Moe Chongqing400044 China College of Computer and Information Science Southwest University Chongqing400715 China

Some studies based on the joint modeling of users’ long- and short-term preferences have achieved significant performance improvements in the next POI prediction task. However, these methods fail to fully explore the potential behavioral commonalities within the long- and short-term visit sequences, thus failing to effectively mitigate the sparsity issue in check-in data. Meanwhile, due to the lack of labeled data that clearly distinguishes general tastes from short-term interests, existing joint modeling approaches for long- and short-term preferences struggle to provide clear supervision and differentiation for target preferences, leading to potential entanglement between these two preferences and consequently affecting the accuracy of preference modeling. Based on these, we propose a novel Disentangling Long- and Short-term Preferences Based on Latent Commonalities Enhancement for Next POI Recommendation (DLSPRec) method to accurately distinguish the differences and commonalities between users’ general tastes and short-term preferences, thereby improving the accuracy of the next location recommendation. This is the first attempt to disentangle long- and short-term preferences in the field of next POI recommendation. Through tests on three real-world datasets, we validated the superiority and rationality of the DLSPRec method. Compared to the suboptimal baseline model, DLSPRec achieved an average performance improvement of 10.39%, 9.69% and 16.39% on the PHO, NYC, and SIN datasets, respectively. The code is available at https://***/IsZSS/DLSPRec. © 2025

关键词： Contrastive learning Next POI recommendation Preferences disentanglement

来源：评论

学校读者我要写书评

暂无评论

Preserving Knowledge from the Source Domain for Cross-Domain Person Re-Identification

SSRN

引用

SSRN 2024年

作者： Ge, Yongxin Gou, Yifeng Li, Ziqiang Zhang, Junyin School of Bigdata and Software Engineering Chongqing University Daxuecheng South Rd. 55 Chognqing400044 China

Although recent cross-domain person re-identification approaches have obtained great progress, they still suffer from two core issues. The first one is the inappropriate immigration of the source domain knowledge. Concretely, this knowledge is not distinguished before being transferred, however, thus the domain-specific knowledge which is detrimental to the target domain performance also be transferred. The second problem is insufficient useful knowledge transfer, which means the beneficial knowledge learned from the source domain is not utilized fully due to the fine-tuning process of the two-stage training especially. To circumvent them, we design a novel collaborative learning method named Preserving Knowledge from the Source Domain (PKSD) from both instance and pixel levels, composed of Ranking-guided Instance Selection (RIS) and Projection based Gradient Selection (PGS). Firstly, the collaborative learning manner could safeguard sufficient knowledge transfer from the source domain. Additionally, RIS tries to select reliable and informative samples from the source domain dataset for training to provide only domain-shared knowledge at the instance level. Subsequently, PGS fine-tunes the feature maps of the selected samples according to the gradient modifying at the pixel level of feature maps to suppress domain-specific knowledge from the source domain. Experiments show that PKSD outperforms existing state-of-the-art methods. © 2024, The Authors. All rights reserved.

关键词： Domain Knowledge

来源：评论

学校读者我要写书评

暂无评论

Preserving Knowledge from the Source Domain for Cross-Domain Person Re-Identification

SSRN

引用

SSRN 2024年

作者： Gou, Yifeng Li, Ziqiang Zhang, Junyin Ge, Yongxin School of Bigdata and Software Engineering Chongqing University Daxuecheng South Rd. 55 Chognqing400044 China

Although recent cross-domain person re-identification approaches have obtained great progress, they still suffer from two core issues. The first one is the inappropriate immigration of the source-domain knowledge. Concretely, this knowledge is not distinguished before being transferred, however, thus the domain-specific knowledge which is detrimental to the target domain performance also be transferred. The second problem is insufficient useful knowledge transfer, which means the beneficial knowledge learned from the source domain is not utilized fully due to the fine-tuning process of the two-stage training especially. To circumvent them, we design a novel collaborative learning method named Preserving Knowledge from the Source Domain (PKSD) from both instance and pixel levels, composed of Ranking-guided Instance Selection (RIS) and Projection based Gradient Selection (PGS). Firstly, the collaborative learning manner could safeguard sufficient knowledge transfer from the source domain. Additionally, RIS tries to select reliable and informative samples from the source-domain dataset for training to provide only domain-shared knowledge at the instance level. Subsequently, PGS fine-tunes the feature maps of the selected samples according to the gradient modifying at the pixel level of feature maps to suppress domain-specific knowledge from the source domain. Experiments show that PKSD outperforms existing state-of-the-art methods. © 2024, The Authors. All rights reserved.

关键词： Domain Knowledge

来源：评论

学校读者我要写书评

暂无评论

Adversarial Classification Rumor Detection based on Social Communication Networks and Time Series Features

Adversarial Classification Rumor Detection based on Social C...

引用

Asia-Pacific Conference on software engineering

作者： Xinyu Zhang Zixin Chang Li Li Wei Zhou Junhao Wen School of Philosophy Sun Yat-Sen University Guangzhou China School of Bigdata and Software Engineering Chongqing University Chongqing China

ISBN: (数字)9798331534011

ISBN: (纸本)9798331534028

With the unprecedented power of social media and public opinion, the widespread rumors on social media have brought a severe negative impact on public life. This paper introduces a new rumor detection method STAC(Spatial-Temporal Adversarial Classifier). STAC aggregates rumor propagation at a finer granularity by jointly modeling their spatial structural and time-series features. In addition, STAC uses an auxiliary adversarial topic classifier to improve generalization to emerging news topics. Experimental results show that STAC exhibits advantages in early detection and overall performance. STAC employs a graph convolutional network to capture the structural information of the propagation graph and uses a bidirectional recurrent neural network to learn the sequence features of the propagation path. Through a gating mechanism, STAC aggregates the learned structural features with temporal features in a fine-grained manner. STAC proves its effectiveness in the experiments by showing advantages in early detection and overall performance. STAC performs better in rumor detection tasks and can detect rumors at an early stage.

关键词： Recurrent neural networks Social networking (online) Graph convolutional networks Aggregates Collaborative filtering Time series analysis Contrastive learning Feature extraction Communication networks software engineering

来源：评论

学校读者我要写书评

暂无评论

Leveraging GCN and User Friendship for Next POI Recommendation 22

Leveraging GCN and User Friendship for Next POI Recommendati...

引用

Proceedings of the 2022 5th International Conference on Machine Learning and Natural Language Processing

作者： Cheng Fu Wei Zhou Junhao Wen Xiangxing Guo Yuhang Chang School of Bigdata and Software Engineering Chongqing University China and Key Laboratory of Dependable Service Computing in Cyber Physical Society Moe China

ISBN: (纸本)9781450399067

The next POI (point of interest) recommendation aims to explore the behaviour patterns from users’ historical check-in records and recommend the next location. However, solving the data sparsity problem and improving the model recommendation performance are still considerable challenges in POI recommendations. This paper proposes a GFUC model consisting of the GCN (Graph Convolutional Networks) and FUC (Friendship of User Context) modules. The GCN module integrates weighted graph convolutional networks to obtain the best representation of users and POIs. The FUC module divides user check-in records with multiple fine-grain manners. Then it incorporates the user friendship and context information, significantly improving the model’s performance while solving the data sparsity problem. More specifically, experimental results show that our proposed GFUC model improves the performance of POI recommendations by more than 10% on both Yelp and Gowalla datasets with the evaluation metrics Precision@10.

关键词： Collaborative Filtering

来源：评论

学校读者我要写书评

暂无评论

Lane marking detection algorithm based on high-precision map and multisensor fusion

Lane marking detection algorithm based on high-precision map...

引用

作者： Yao, Haichang Chen, Chen Liu, Shangdong Li, Kui Ji, Yimu Huang, Guangyan Wang, Ruchuan School of Computer Science Nanjing University of Posts and Telecommunications Nanjing China School of Computer and Software Nanjing Institute of Industry Technology Nanjing China Industrial Software Engineering Technology Research and Development Center of Jiangsu Education Department Nanjing Institute of Industry Technology Nanjing China Jiangsu HPC and Intelligent Processing Engineer Research Center Nanjing China Institute of High Performance Computing and Bigdata Nanjing University of Posts and Telecommunications Nanjing China School of Information Technology Deakin University Melbourne Australia

In case of sharp road illumination changes, bad weather such as rain, snow or fog, wear or missing of the lane marking, the reflective water stain on the road surface, the shadow obstruction of the tree, and mixed lane markings and other signs, missing detection or wrong detection will occur for the traditional lane marking detection algorithm. In this manuscript, a lane marking detection algorithm based on high-precision map and multisensor fusion is proposed. The basic principle of the algorithm is to use the centimeter-level high-precision positioning combined with high-precision map data to complete the detection of lane markings. In the process of generating high-precision maps or in the uncovered areas of high-precision maps, LIDAR (LIght Detection And Ranging) is used to estimate the curvature of the road to assist in lane marking detection. The experimental results show that the algorithm has lower false detection rate in case of bad road conditions, and the algorithm is robust. © 2020 John Wiley & Sons, Ltd.

关键词： Roads and streets

来源：评论

学校读者我要写书评

暂无评论

VulDetectBench: Evaluating the Deep Capability of Vulnerability Detection with Large Language Models

arXiv

引用

arXiv 2024年

作者： Liu, Yu Gao, Lang Yang, Mingxin Xie, Yu Chen, Ping Zhang, Xiaojin Chen, Wei Institute of BigData Fudan University Shanghai China School of Computer Science Huazhong University of Science and Technology China Purple Mountain Laboratories Nanjing China School of Computer Science Fudan University Shanghai China School of Software Engineering Huazhong University of Science and Technology China

Large Language Models (LLMs) have training corpora containing large amounts of program code, greatly improving the model’s code comprehension and generation capabilities. However, sound comprehensive research on detecting program vulnerabilities, a more specific task related to code, and evaluating the performance of LLMs in this more specialized scenario is still lacking. To address common challenges in vulnerability analysis, our study introduces a new benchmark, VulDetectBench, specifically designed to assess the vulnerability detection capabilities of LLMs. The benchmark comprehensively evaluates LLM’s ability to identify, classify, and locate vulnerabilities through five tasks of increasing difficulty. We evaluate the performance of 17 models (both open- and closed-source) and find that while existing models can achieve over 80% accuracy on tasks related to vulnerability identification and classification, they still fall short on specific, more detailed vulnerability analysis tasks, with less than 30% accuracy, making it difficult to provide valuable auxiliary information for professional vulnerability mining. Our benchmark effectively evaluates the capabilities of various LLMs at different levels in the specific task of vulnerability detection, providing a foundation for future research and improvements in this critical area of code security. VulDetectBench is publicly available at https://***/Sweetaroo/VulDetectBench. © 2024, CC BY-NC-SA.

关键词： Classification (of information)

来源：评论

学校读者我要写书评

暂无评论

Enhanced Contrastive Learning with Multi-Aspect Information for Recommender Systems

SSRN

引用

SSRN 2022年

作者： Hu, Linfeng Zhou, Wei Luo, Fengji Ni, Shuang Wen, Junhao School of Bigdata and Software Engineering Chongqing University Daxuecheng South Rd. 55 Chognqing400044 China Key Laboratory of Dependable Service Computing in Cyber Physical Society MOE Chognqing400044 China School of Civil Engineering The University of Sydney SydneyNSW2006 Australia

In recent years, graph neural networks (GNNs) as the state-of-art have advanced the solutions of recommender systems. However, the embedding representations of nodes obtained from GNNs-based models are always suboptimal when facing sparse interaction data. Contrastive learning on graphs is proven effective in alleviating the influence of data sparsity. Most related works perform contrastive learning through self-discrimination while ignoring the self-supervised signals from neighboring nodes, which can also be quite crucial in optimizing the representations. To tackle these issues, we propose a model named Enhanced Contrast Learning with Multi-Aspect Information (ECMI) for recommender systems, which aims to capture the self-supervised signals from both the nodes themselves and their neighbors. Specifically, we conducted contrastive learning in the two-layer space separately. We consider structurally directly connected homophilic node pairs as positive examples in explicit space while introducing a social relationship network and an item entity network in implicit space. These two networks can be combined with the interaction network to conduct a two-channel tri-training procedure. This procedure can build more effective contrastive learning by exploring the similar potential neighbors of the target node. We conduct experiments on three public datasets, and the results consistently validate the effectiveness of ECMI. © 2022, The Authors. All rights reserved.

关键词： Recommender systems

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：