检索结果-内蒙古大学图书馆

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Hao Wang Xiaochuan Wang Ruijun Liu Xiaoming Chen Haisheng Li School of Computer and Artificial Intelligence Beijing Technology and Business University School of Software Engineering Beijing University

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

Recent transformer-based methods have achieved notable success in 3D human pose estimation. However, the most utilized self-attention mechanisms compute the attention matrix by performing a dot product on inter-vector features, which may overlook finer element-wise interactions. In this paper, we introduce star-attention, which integrates the star operation into the self-attention module. This approach retains the non-linearity and high dimensionality characteristics of matrix multiplication in self-attention while enhancing the feature representation by capturing interactions at a finer granularity. Specifically, we validated the promotion of the proposed star-attention module on MixSTE as an example. Experimental results demonstrate that our approach achieves competitive performance on benchmarks compared to state-of-the-art methods, producing smoother 3D poses across successive frames.

关键词： Three-dimensional displays Pose estimation Stars Benchmark testing Signal processing Transformers Acoustics Speech processing

来源：评论

学校读者我要写书评

暂无评论

Preview-based Category Contrastive Learning for Knowledge Distillation

引用

IEEE Transactions on Circuits and Systems for Video technology 2025年

作者： Ding, Muhe Wu, Jianlong Dong, Xue Li, Xiaojie Qin, Pengda Gan, Tian Nie, Liqiang School of Computer Science and Technology 518055 China Tsinghua University School of Software Beijing100084 China Alibaba Group Beijing100102 China Shandong University School of Computer Science and Technology Qingdao266237 China

Knowledge distillation is a mainstream algorithm in model compression by transferring knowledge from the larger model (teacher) to the smaller model (student) to improve the performance of student. Despite many efforts, existing methods mainly investigate the consistency between instance-level feature representation or prediction, which neglects the category-level information and the difficulty of each sample, leading to undesirable performance. To address these issues, we propose a novel preview-based category contrastive learning method for knowledge distillation (PCKD). It first distills the structural knowledge of both instance-level feature correspondence and the relation between instance features and category centers in a contrastive learning fashion, which can explicitly optimize the category representation and explore the distinct correlation between representations of instances and categories, contributing to discriminative category centers and better classification results. Besides, we introduce a novel preview strategy to dynamically determine how much the student should learn from each sample according to their difficulty. Different from existing methods that treat all samples equally and curriculum learning that simply filters out hard samples, our method assigns a small weight for hard instances as a preview to better guide the student training. Extensive experiments on several challenging datasets, including CIFAR-100, ImageNet and Pascal VOC, demonstrate the superiority over state-of-the-art methods. © 1991-2012 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Random Masked Autoencoders Framework for Multi-View Multi-Label Learning with Incomplete Views and Labels 25

Random Masked Autoencoders Framework for Multi-View Multi-La...

引用

Proceedings of the 2025 5th International Conference on Bioinformatics and Intelligent Computing

作者： Bo Cui Qing Ai Yuting Xu Yiying Wang School of Computer Science and Software Engineering University of Science and Technology Liaoning Anshan China

ISBN: (纸本)9798400712203

Dual incomplete multi-view multi-label learning (DIMVMLL) aims to make accurate predictions for multiple labels by exploiting incomplete multi-view information and incomplete multi-label information from input samples. Since the view information and label information of samples are incomplete in this task, existing DIMVMLL methods classify samples based on the semantic relationships between view features and labels. However, in the DIMVMLL task, exploring the consistent and complementary information between views is important for the classification performance of model. To learn high-quality view consistency information and complementary information, we propose random masked autoencoders framework (RMAF). First, we design a triple reconstruction loss for the dual-channel autoencoders to guide the model in extracting consistent and complementary information between views and introduce contrastive learning to guide the model in learning private information for each view. Second, we propose a feature-level random masking strategy to enhance high-level feature extract ability of the autoencoders. Furthermore, to preserve the feature semantic structure, we introduce structure loss. Extensive experiments have validated the effectiveness of our model.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

A Universal Framework for Reversible Data Hiding in Encrypted Images with Multiple Hiders

引用

IEEE Transactions on Dependable and Secure Computing 2025年

作者： Xiong, Lizhi Zhou, Wenhao Xia, Zhihua Weng, Jian School of Computer and Software Nanjing University of Information science and Technology Nanjing China Jinan University Guangzhou China

Recently, Reversible Data Hiding in Encrypted Images with Multiple Hiders (RDHEI-MH) has attracted the attention of researchers, as it can satisfy the requirements of multiparty embedding. In this paper, a universal framework for RDHEI-MH is proposed, in which most of existing RDH algorithms in the plaintext domain (including Data Expansion (DE)-based and Histogram Shifting (HS)-based algorithms) can be applied in encrypted images without performance loss. That cannot be achieved by current existing schemes. The framework contains three entities: image owner, multiple hiders, and image receiver. The cover image is shared by Additive Secret Sharing (ASS) at the image owner’s side, and then each image share is sent to the corresponding hider. Based on the Secure Multiparty Computation (MPC) protocols specially designed for RDH operations, the hiders can securely perform computation on shares to embed additional data. After receiving all the marked image shares, the image receiver can reconstruct the marked image, and further extract the data. Compared with the previous RDHEI-MH schemes, the proposed scheme achieves full-separability, i.e., the extraction and decryption operations are commutative. The experimental results and theoretical analyses demonstrate that the proposed scheme has high visual quality of the reconstructed image and high embedding capacity, while its security is proved. © 2004-2012 IEEE All rights reserved.

关键词： Additive secret sharing Reversible data hiding in encrypted image Secure multiparty computation Secure protocol

来源：评论

学校读者我要写书评

暂无评论

An Improved LVSL with View Consensus 25

An Improved LVSL with View Consensus

引用

Proceedings of the 2025 5th International Conference on Bioinformatics and Intelligent Computing

作者： Yuting Xu Qing Ai Bo Cui School of Computer Science and Software Engineering University of Science and Technology Liaoning Anshan China

ISBN: (纸本)9798400712203

In multi-view multi-label learning (MVML), each sample can be represented by multiple view features and associated with multiple labels. Most existing MVML algorithms are based on the assumption that all views share the same label set. However, in practice, different views may contain distinct label information, that means a single view cannot fully represent all labels. Based on this issue, the LVSL algorithm effectively learns view-specific labels and obtains superior classification performance. However, LVSL still has the limitation that it fails to consider the correlations between views, leading to suboptimal learning results. In this paper, we propose an improved LVSL algorithm named LVSL_VC (LVSL with view consensus). We incorporate view consensus learning into the original LVSL framework. Firstly, we employ view weights to model view consensus, assuming that views with similar weights will yield similar prediction outputs, conversely, they will be different. Secondly, we integrate the view consensus into the LVSL framework and construct a new classification model. Finally, we utilize an alternating optimization method to solve the problem. Extensive experimental results demonstrate that the LVSL_VC outperforms other state-of-the-art MVML algorithms.

关键词： Multi-view multi-label learning

来源：评论

学校读者我要写书评

暂无评论

Tensorized latent representation with automatic dimensionality selection for multi-view clustering

引用

Pattern Recognition 2025年 160卷

作者： Cai, Bing Lu, Gui-Fu Guo, Xiaoxing Wu, Tong School of Computer and Software Engineering Anhui Institute of Information Technology Wuhu241000 China School of Computer Science and Engineering Nanjing University of Science and Technology Nanjing210094 China School of Computer and Information Anhui Polytechnic University Wuhu241000 China

Latent representation has garnered significant attention in the field of multi-view learning due to its ability to capture the underlying structures of raw data and achieve promising results. However, latent representation-based methods often encounter challenges in selecting the dimensionality of the latent view, which limits their applicability. To address this problem, we propose a novel method called Tensorized Latent Representation with Automatic Dimensionality Selection (TLRADS), which can automatically determine the optimal dimensions. In TLRADS, we leverage the cumulative contribution rate of singular values to determine the number of dimensions for each view-specific latent representation. This approach ensures that the chosen dimensions capture a significant portion of the data's variability while discarding less relevant information. After obtaining the latent representation views, we incorporate the tensor subspace learning technique to capture the underlying structural information more comprehensively. Finally, an efficient iterative algorithm is designed to solve the TLRADS model. Through experimental validation, we demonstrate the effectiveness of the automatic dimensionality selection strategy in TLRADS. Meanwhile, the experimental results on real-life datasets indicate that TLRADS outperforms state-of-the-art multi-view clustering methods. © 2024 Elsevier Ltd

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

SLAM: Towards Efficient Multilingual Reasoning via Selective Language Alignment 31

SLAM: Towards Efficient Multilingual Reasoning via Selective...

引用

31st International Conference on Computational Linguistics, COLING 2025

作者： Fan, Yuchun Mu, Yongyu Wang, Yilin Huang, Lei Ruan, Junhao Li, Bei Xiao, Tong Huang, Shujian Feng, Xiaocheng Zhu, Jingbo NLP Lab School of Computer Science and Engineering Northeastern University Shenyang China NiuTrans Research Shenyang China Harbin Institute of Technology Harbin China Meituan Inc. National Key Laboratory for Novel Software Technology Nanjing University China

ISBN: (纸本)9798891761964

Despite the significant improvements achieved by large language models (LLMs) in English reasoning tasks, these models continue to struggle with multilingual reasoning. Recent studies leverage a full-parameter and two-stage training paradigm to teach models to first understand non-English questions and then reason. However, this method suffers from both substantial computational resource computing and catastrophic forgetting. The fundamental cause is that, with the primary goal of enhancing multilingual comprehension, an excessive number of irrelevant layers and parameters are tuned during the first stage. Given our findings that the representation learning of languages is merely conducted in lower-level layers, we propose an efficient multilingual reasoning alignment approach that precisely identifies and fine-tunes the layers responsible for handling multilingualism. Experimental results show that our method, SLAM, only tunes 6 layers' feed-forward sub-layers including 6.5−8% of all parameters within 7B and 13B LLMs, achieving superior average performance than all strong baselines across 10 languages. Meanwhile, SLAM only involves one training stage, reducing training time by 4.1 − 11.9× compared to the two-stage method. © 2025 Association for Computational Linguistics.

关键词： Computational linguistics

来源：评论

学校读者我要写书评

暂无评论

Hard Sample Aware Robust Contrastive Learning for Multi-View Clustering

Hard Sample Aware Robust Contrastive Learning for Multi-View...

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Yuanzhe Cai Zhikui Chen Jing Gao Peng Li Jianing Zhang School of Software Dalian University of Technology Dalian China School of Computer Science and Technolgoy Dalian University of Technology Dalian China

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

Multi-view clustering aims to divide samples into several clusters, by mining and utilizing the consistency and complementarity of multi-view data. Recent years, numerous deep contrastive multi-view clustering methods have been proposed to address the false negative issue by using self-supervised information. However, the quality of these self-supervised information was rarely taken into consideration, and using these information without discrimination can compromise training, leading to sub optimal performance. To tackle this issue, we propose Hard Sample Aware Robust Contrastive Learning for Multi-View Clustering(HearMVC). Concretely, we use self-supervised information and similarity to determine hard samples. The model focuses on these hard samples by assigning higher weights to enhance discriminative capability. Moreover, we utilize the confidence of self-supervised cluster assignment as weights, to strengthen the learning to confident samples and weaken the influence of unconfident samples. By simultaneously considering the weighting of hardness and confidence, our method can achieve best robustness and strongest discriminative capability. Extensive experiments on public datesets verify the effectiveness of our method.

关键词： Training Clustering methods Contrastive learning Signal processing Robustness Acoustics Data mining Speech processing

来源：评论

学校读者我要写书评

暂无评论

Adaptive Occlusion Face Recognition Feature Fusion Network Based on Precise Guidance Mask Extraction

引用

Journal of Shanghai Jiaotong University (Science) 2025年 1-12页

作者： Ye, Jihua Zhang, Lei Wang, Chao Tu, Qixuan Zheng, Rehua Huang, Huiyuan Jiang, Aiwen School of Computer and Information Engineering Jiangxi Normal University Nanchang330022 China Jiangxi Provincial Key Laboratory for High Performance Computing Jiangxi Normal University Nanchang330022 China State International Science & Technology Cooperation Base of Networked Supporting Software Jiangxi Normal University Nanchang330022 China

The uncertainty in the position and size of occluding objects greatly affects the extraction of identity features in facial recognition, which is a challenge that existing methods fail to effectively address. To tackle this issue, this paper introduces a novel feature fusion network for occluded facial recognition that adaptively handles arbitrary occlusions and accurately extracts occlusion information masks. The network effectively mitigates the impact of occlusions on facial identity features through a feature fusion module. The proposed method employs an occlusion classification module and an occlusion classification loss to precisely guide the extraction of accurate occlusion masks. Subsequently, an occlusion-aware hybrid attention module is introduced, which focuses on the relationship between facial features and occlusion masks across channel and spatial dimensions. This fusion strategy enables the extraction of facial identity features that are resilient to occlusions. Experimental results indicate that the proposed method holds a significant advantage in handling facial recognition tasks with arbitrary occlusions. On the LFW and MF1 synthetic occlusion datasets, the average accuracy rates reached 93.19% and 72.57%, respectively, showing improvements of 1.7 and 2.39 percent points over mainstream methods. © Shanghai Jiao Tong University 2025.

关键词： A feature fusion occluded facial recognition occlusion mask occlusion perception TP391.41

来源：评论

学校读者我要写书评

暂无评论

DA-FR-UNet: A Dual Attention-Enhanced Approach for Retinal Vessel Image Segmentation

IAENG International Journal of Computer Science

引用

IAENG International Journal of computer Science 2025年第5期52卷 1325-1335页

作者： Li, Yuanyuan Wu, Jiansheng Liu, Yang School of Computer Science and Software Engineering University of Science and Technology Liaoning Anshan114051 China University of Science and Technology Liaoning Anshan114051 China

Retinal vessel image segmentation is crucial for the early diagnosis and treatment of ophthalmic diseases. However, accurate segmentation remains challenging due to the complexity and diversity of retinal images, as well as the intricate nature of vascular structures. Existing methods often underperform when dealing with fine vessels and complex backgrounds, limiting their effectiveness in clinical applications. This study proposes a novel retinal vessel image segmentation method, Dual Attention FR-UNet (DA-FR-UNet), which incorporates two innovative attention mechanisms: Enhanced Channel Prior Attention (ECPA) and Hard Triplet Attention (HTA). The ECPA module enhances the model's ability to recognize fine blood vessels by integrating global average pooling and global maximum pooling to effectively capture multi-scale features. In contrast, the HTA module improves the model's understanding of complex vascular structures by simultaneously considering attention information across three dimensions: channel, spatial height, and width. Additionally, the study introduces the testtime augmentation (TTA) technique during the experimental phase to enhance the model's generalizability and robustness. To validate the performance of the DA-FR-UNet model in retinal vessel image segmentation tasks, a series of comparative experiments were conducted on the FIVES and RVD datasets. The experimental results demonstrate that the DA-FR-UNet model, combining the dual-attention mechanism and the TTA strategy, outperforms existing state-of-the-art methods across several key evaluation metrics. On the FIVES dataset, DAFR- UNet achieves a sensitivity of 88.42%, an IoU value of 82.27%, and a Dice coefficient of 89.72%. On the RVD dataset, these metrics are 65.46%, 44.53%, and 61.43%, respectively. These findings highlight the model's significant advantages in processing fine vessels and complex backgrounds, improving its sensitivity and robustness in blood vessel image segmentation. © (2025)

关键词： Image segmentation

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：