检索结果-内蒙古大学图书馆

arXiv 2024年

作者： Chen, Geng Yang, Junqing Pu, Xiaozhou Ji, Ge-Peng Xiong, Huan Pan, Yongsheng Cui, Hengfei Xia, Yong National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology School of Computer Science and Engineering Northwestern Polytechnical University Xi’an China School of Computing Australian National University Canberra Australia Mohamed bin Zayed University of Artificial Intelligence United Arab Emirates Harbin Institute of Technology China

Accurate segmentation of polyps from colonoscopy videos is of great significance to polyp treatment and early prevention of colorectal cancer. However, it is challenging due to the difficulties associated with modelling long-range spatio-temporal relationships within a colonoscopy video. In this paper, we address this challenging task with a novel Mixture-Attention Siamese Transformer (MAST), which explicitly models the long-range spatio-temporal relationships with a mixture-attention mechanism for accurate polyp segmentation. Specifically, we first construct a Siamese transformer architecture to jointly encode paired video frames for their feature representations. We then design a mixture-attention module to exploit the intra-frame and inter-frame correlations, enhancing the features with rich spatio-temporal relationships. Finally, the enhanced features are fed to two parallel decoders for predicting the segmentation maps. To the best of our knowledge, our MAST is the first transformer model dedicated to video polyp segmentation. Extensive experiments on the large-scale SUN-SEG benchmark demonstrate the superior performance of MAST in comparison with the cutting-edge competitors. Our code is publicly available at https://***/Junqing-Yang/MAST. Copyright © 2024, The Authors. All rights reserved.

关键词： Benchmarking

来源：评论

学校读者我要写书评

暂无评论

Keeping Pace with Ever-Increasing data: Towards Continual Learning of Code Intelligence Models

Keeping Pace with Ever-Increasing Data: Towards Continual Le...

引用

International Conference on Software engineering (ICSE)

作者： Shuzheng Gao Hongyu Zhang Cuiyun Gao Chaozheng Wang School of Computer Science and Technology Harbin Institute of Technology Shenzhen China Peng Cheng Laboratory and Guangdong Provincial Key Laboratory of Novel Security Intelligence Technologies School of Big Data and Software Engineering Chongqing University China

Previous research on code intelligence usually trains a deep learning model on a fixed dataset in an offline manner. However, in real-world scenarios, new code repositories emerge incessantly, and the carried new knowledge is beneficial for providing up-to-date code intelligence services to developers. In this paper, we aim at the following problem: How to enable code intelligence models to continually learn from ever-increasing data? One major challenge here is catastrophic forgetting, meaning that the model can easily forget knowledge learned from previous datasets when learning from the new dataset. To tackle this challenge, we propose REPEAT, a novel method for continual learning of code intelligence models. Specifically, REPEAT addresses the catastrophic forgetting problem with representative exemplars replay and adaptive parameter regularization. The representative exemplars replay component selects informative and diverse exemplars in each dataset and uses them to re-train model periodically. The adaptive parameter regularization component recognizes important parameters in the model and adaptively penalizes their changes to preserve the knowledge learned before. We evaluate the proposed approach on three code intelligence tasks including code summarization, software vulnerability detection, and code clone detection. Extensive experiments demonstrate that REPEAT consistently outperforms baseline methods on all tasks. For example, REPEAT improves the conventional fine-tuning method by 1.22, 5.61, and 1.72 on code summarization, vulnerability detection and clone detection, respectively.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Epl: Empirical Prototype Learning for Deep Face Recognition

SSRN

引用

SSRN 2024年

作者： Fan, Weijia Wen, Jiajun Jia, Xi Shen, Linlin Zhou, Jiancan Li, Qiufu National Engineering Laboratory for Big Data System Computing Technology Shenzhen University Shenzhen518060 China Computer Vision Institute Shenzhen University Shenzhen518060 China School of Computer Science University of Birmingham Birmingham United Kingdom Aqara Lumi United Technology Coopration Shenzhen China

Prototype learning is widely used in face recognition, which takes the row vectors of coefficient matrix in the last linear layer of the feature extraction model as the prototypes for each class. When the prototypes are updated using the facial sample feature gradients in the model training, they are prone to being pulled away from the class center by the hard samples, resulting in decreased overall model performance. this paper, we explicitly define prototypes as the expectations of sample features in each class and design the empirical prototypes using the existing samples in the dataset. We then devise a strategy to adaptively update these empirical prototypes during the model training based on the similarity between the sample features and the empirical prototypes. Furthermore, we propose an empirical prototype learning (EPL) method, which utilizes an adaptive margin parameter with respect to sample features. EPL assigns larger margins to the normal samples and smaller margins to the hard samples, allowing the learned empirical prototypes to better reflect the class center dominated by the normal samples and finally pull the hard samples towards the empirical prototypes through the learning. extensive experiments on MFR, IJB-C, LFW, CFP-FP, AgeDB, and MegaFace demonstrate the effectiveness of EPL. Our code is available at https://***/WakingHours-GitHub/EPL. © 2024, The Authors. All rights reserved.

关键词： Face recognition

来源：评论

学校读者我要写书评

暂无评论

DE-ConvGraph 3D UNet: A Novel Deep Learning Model for Optimizing Radiotherapy Treatment Plans in Oropharyngeal Cancer

DE-ConvGraph 3D UNet: A Novel Deep Learning Model for Optimi...

引用

Applied Imagery Pattern Recognition Workshop (AIPR)

作者： Bhuvanashree Murugadoss J Amudha Vijayan Sugumaran Department of Computer Science and Engineering School of Computing Amrita Vishwa Vidyapeetham Bengaluru Karnataka India Center for Data Science and Big Data Analytics Oakland University Rochester MI USA Department of Decision and Information Sciences Oakland University Rochester MI USA

Automated radiotherapy treatment planning aims to improve treatment accuracy and efficiency. However, the prevalent Knowledge-Based Planning (KBP) method faces issues like lengthy manual problem formulation and challenges in accurately modeling human anatomy and processing high-dimensional data. This work focuses on treatment plan optimization and considers deep learning as a potential solution. Deep learning algorithms, by virtue of their ability to learn from large quantities of data and model complex relationships, can automate the formulation of the optimization problem in KBP, saving significant time and effort. However, despite these compelling advantages, it should be noted that the current convolution-based encoder-decoder models used for radiotherapy treatment plan optimization have a limited capability in capturing long-range dependencies between distant voxels. This work aims to introduce ‘DE-ConvGraph 3D UNet’, a novel deep learning model, to address these limitations and optimize radiotherapy treatment plans for oropharyngeal cancer. The proposed ‘DE-ConvGraph 3D UNet’ model includes a Graph Convolutional Network (GCN) component to capture long-range dependencies between distant voxels. Furthermore, the dual-encoder structure of the model combines the strengths of GCN and 3D Convolutional U-Net, enabling global relationships and local pattern capturing in a 3D patient volume. Experiments were conducted using the benchmark dataset - OpenKBP-Opt. The proposed model shows improvements in performance in comparison to state-of-the-art U-Net variants in terms of mean squared voxel-wise error, dose volume histogram points and clinical criteria satisfaction.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Combating Misinformation/ Disinformation in Online Social Media: A Multidisciplinary View

Combating Misinformation/ Disinformation in Online Social Me...

引用

作者： Barni, Mauro Fang, Yi Liu, Yuhong Robinson, Laura Sasahara, Kazutoshi Vincent, Subramaniam Wang, Xinchao Wu, Zhizheng Department of Information Engineering and Mathematics University of Siena Italy Department of Computer Science and Engineering Santa Clara University United States Department of Sociology Santa Clara University United States Department of Innovation Science Tokyo Institute of Technology Japan Markkula Center for Applied Ethics Santa Clara University United States Department of Electrical and Computer Engineering National University of Singapore Singapore School of Data Science Chinese Univ. of Hong Kong Shenzhen China

Recently, the viral propagation of mis/disinformation has raised significant concerns from both academia and industry. This problem is particularly difficult because on the one hand, rapidly evolving technology makes it much cheaper and easier to manipulate and propagate social media information. On the other hand, the complexity of human psychology and sociology makes the understanding, prediction and prevention of users' involvement in mis/disinformation propagation very difficult. This themed series on "Multi-Disciplinary Dis/Misinformation Analysis and Countermeasures"aims to bring the attention and efforts from researchers in relevant disciplines together to tackle this challenging problem. In addition, on October 20th, 2021, and March 7th 2022, some of the guest editorial team members organized two panel discussions on "Social Media Disinformation and its Impact on Public Health During the COVID-19 Pandemic,"and on "Dis/Misinformation Analysis and Countermeasures - A Computational Viewpoint."This article summarizes the key discussion items at these two panels and hopes to shed light on the future directions. © 2022 M. Barni, Y. Fang, Y. Liu, L. Robinson, K. Sasahara, S. Vincent, X. Wang and Z. Wu.

关键词： Social networking (online)

来源：评论

学校读者我要写书评

暂无评论

Detection of Pre Shoplifting Suspicious Behavior Using Deep Learning

Detection of Pre Shoplifting Suspicious Behavior Using Deep ...

引用

IIAI International Conference on Advanced Applied Informatics (IIAIAAI)

作者： Sujan Shrestha Yoji Taniguchi Tetsuo Tanaka Course of Information and computer sciences Graduate school of Engineering Kanagawa Institute of Tehnology Atsugi Japan Department of Information Artificial Intelligence and Data Science Daiichi Institute of Technology Tokyo Japan Dept. of Information and Computer Sciences Faculty of Information Technology Kanagawa Institute of Technology Atsugi Japan

ISBN: (数字)9798350377903

ISBN: (纸本)9798350377910

Shoplifting presents a significant challenge for retailers, leading to substantial financial losses and security concerns. Current surveillance systems effectively identify shoplifting incidents postoccurrence; however, the capability to detect and prevent such activities preemptively remains limited. This research focuses on identifying behaviors that likely precede shoplifting events. To this end, we introduce a novel benchmark dataset, specifically designed for the early detection of suspicious behavior that may lead to shoplifting. This dataset is curated by selecting and segmenting relevant sequences from the UCF101-Crime dataset, encompassing suspicious behavior, actual shoplifting incidents, and normal activities. Our approach addresses the limitations of existing datasets where the brief duration of shoplifting events may skew neural network performance. The dataset includes clips of only a few seconds each, for which we propose a robust neural network architecture combining the strengths of convolutional neural networks (CNN) for efficient feature extraction and Bidirectional long short-term memory (BiLSTM) networks for capturing temporal dependencies within action sequences. Our model achieves an impressive classification accuracy of 86.21 % on the newly created dataset.

关键词： Deep learning Recurrent neural networks Accuracy Surveillance Benchmark testing Feature extraction Convolutional neural networks Security Informatics Long short term memory

来源：评论

学校读者我要写书评

暂无评论

Multi-instance learning anomaly event detection based on Transformer 22

Multi-instance learning anomaly event detection based on Tra...

引用

Proceedings of the 2022 5th International Conference on Artificial Intelligence and Pattern Recognition

作者： Feifei Qin Yuelei Xiao School of Computer Xi'an University of Posts and Telecommunications China School of Modern Posts Xi'an University of Posts and Telecommunications China and Big Data Industry Application Research Department Shaanxi Provincial Information Engineering Research Institute China

ISBN: (纸本)9781450396899

Multi-instance learning (MIL) is the dominant approach for weakly supervised anomaly detection in surveillance videos. The shortcomings of using the features extracted by networks such as Convolutional 3D (C3D) or inflated 3D-ConvNet (I3D) alone to extract video context features have prompted the emergence of various abnormal event detection algorithms based on attention mechanisms. Vision Transformer (ViT) applies transformer to the field of computer vision for the first time and demonstrates its superior performance. In this paper, we propose a multi-instance learning anomaly event detection method based on Transformer, called MIL-ViT, which uses an inflated I3D pre-training model to extract Spatio-temporal features, and then inputs features into the ViT encoder to extract the particular salient pieces of information, and the anomaly scores are obtained. Furthermore, we introduce the MIL ranking loss and the center loss function for better training. The experimental results on two benchmark datasets (i.e. ShanghaiTech and UCF-Crime) show that the AUC value of our method is significantly improved compared with several state-of-the-art methods in recent years.

关键词： Anomaly detection multi-instance learning surveillance videos vision transformer

来源：评论

学校读者我要写书评

暂无评论

Tensor Levenberg-Marquardt Algorithm for Multi-Relational Traffic Prediction

引用

IEEE Transactions on Vehicular Technology 2023年第9期72卷 11275-11290页

作者： Chang, Shih Yu Wu, Hsiao-Chun Kuan, Yen-Cheng Wu, Yiyan San Jose State University Department of Applied Data Science San JoseCA95192 United States Louisiana State University School of Electrical Engineering and Computer Science Baton RougeLA70803 United States Yuan Ze University Innovation Center for AI Applications Chungli32003 Taiwan International College of Semiconductor Technology Hsinchu30010 Taiwan Communications Research Centre Canada OttawaONK2H 8S2 Canada

Traffic prediction serves as a crucial role in intelligent transportation systems for applications including routing planning and traffic control. With the advances of sensor technologies and Internet of Things (IoTs), the exploitation of traffic data motivates us to reformulate the traffic-state prediction problem based on the big-data characteristics. In order to infer multifarious traffic information, the standard one/two-relational traffic data in the vector/matrix form needs to be expanded to multi-relational traffic data in an arbitrary tensor form. However, none of the existing approaches is capable of performing traffic prediction by characterizing and tracking the inherent multi-relational nonlinear dynamics which are often encountered in the realistic time-series analysis. Furthermore, there hardly exist convergence studies on convergence condition and convergence speed pertinent to the existing traffic-prediction methods. In order to combat the aforementioned challenges, we propose a new tensor Levenberg-Marquadt algorithm (TLMA) to conduct the multi-relational nonlinear regression for traffic prediction involving tensorized data. Meanwhile, we also derive the lower-bound of the number of iterations required to make the error-norm below a specified tolerance for global convergence. Besides, we derive the conditions for our proposed TLMA to converge to the optimal solution. The convergence rates with respect to different nonlinear cost functions have been derived theoretically. The computational- and memory-complexities of our proposed new TLMA are studied as well. Finally, numerical experiments are conducted to evaluate the traffic prediction performance of our proposed new TLMA over real traffic data in comparison with other existing multi-relational prediction methods. © 1967-2012 IEEE.

关键词： Tensors

来源：评论

学校读者我要写书评

暂无评论

STFormer: Learning to Explore Spot Relationships for Spatial Transcriptomics Prediction from Histology of Colorectal Cancer

STFormer: Learning to Explore Spot Relationships for Spatial...

引用

Annual International Conference of the IEEE engineering in Medicine and Biology Society (EMBC)

作者： Gan Zhan Xiuju Du Jing Liu Yinhao Li Lanfen Lin Jingsong Li Yen-Wei Chen Graduate School of Information Science and Engineering Ritsumeikan University Shiga Japan Research Center for Healthcare Data Science Zhejiang Lab Hangzhou China College of Information Science and Engineering Ritsumeikan University Shiga Japan College of Computer Science and Technology Zhejiang University Hangzhou China

ISBN: (数字)9798350371499

ISBN: (纸本)9798350371505

Spatial transcriptomics (ST) offers insights into gene expression patterns within tumor microenvironments, but its widespread application is impeded by cost constraints. To address this, predicting ST from Histology emerges as a cost-effective alternative. However, current methods such as STNet, HistoGene, and Hist2ST exhibit limitations, either overlooking stain variation across datasets or failing to well explore inter-spot correlations in scenarios with limited Whole Slide Image (WSI) data. In response, we present STFormer, a deep learning approach that incorporates the Style-Aug module to enhance feature generalization through medically-irrelevant style transfer augmentation. Additionally, the Cross-WSI Transformer module is introduced to capture Cross-WSI spot relationships efficiently. Our experimental results, conducted on both internal and external datasets, demonstrate that STFormer surpasses existing methods by a substantial margin. This showcases its potential as a powerful tool for spatial transcriptomics predictions, addressing critical gaps in current methodologies.

关键词： Deep learning Costs Correlation Histopathology Transcriptomics Colorectal cancer Transformers Gene expression Tumors Biomedical imaging

来源：评论

学校读者我要写书评

暂无评论

MetaGCD: Learning to Continually Learn in Generalized Category Discovery

MetaGCD: Learning to Continually Learn in Generalized Catego...

引用

International Conference on computer Vision (ICCV)

作者： Yanan Wu Zhixiang Chi Yang Wang Songhe Feng Key Laboratory of Big Data & Artificial Intelligence in Transportation Ministry of Education Beijing Jiaotong University Beijing China School of Computer and Information Technology Beijing Jiaotong University Beijing China Department of Electrical and Computer Engineering University of Toronto Toronto Canada Department of Computer Science and Software Engineering Concordia University Montreal Canada

In this paper, we consider a real-world scenario where a model that is trained on pre-defined classes continually encounters unlabeled data that contains both known and novel classes. The goal is to continually discover novel classes while maintaining the performance in known classes. We name the setting Continual Generalized Category Discovery (C-GCD). Existing methods for novel class discovery cannot directly handle the C-GCD setting due to some unrealistic assumptions, such as the unlabeled data only containing novel classes. Furthermore, they fail to discover novel classes in a continual fashion. In this work, we lift all these assumptions and propose an approach, called MetaGCD, to learn how to incrementally discover with less forgetting. Our proposed method uses a meta-learning framework and leverages the offline labeled data to simulate the testing incremental learning process. A meta-objective is defined to revolve around two conflicting learning objectives to achieve novel class discovery without forgetting. Furthermore, a soft neighborhood-based contrastive network is proposed to discriminate uncorrelated images while attracting correlated images. We build strong baselines and conduct extensive experiments on three widely used benchmarks to demonstrate the superiority of our method.

关键词：

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：