检索结果-内蒙古大学图书馆

Transformers for Vision: A Survey on Innovative Methods for computer Vision

IEEE Access 2025年 13卷 95496-95523页

作者： Hassija, Vikas Palanisamy, Balamurugan Chatterjee, Arpita Mandal, Arpita Chakraborty, Debanshi Pandey, Amit Chalapathi, Gss Kumar, Dhruv Deemed to be University School of Computer Engineering Odisha Bhubaneswar751024 India Bennett University School of CSET Gautam Buddha Nagar Greater Noida201310 India Department of Electrical and Electronics Engineering Birla Institute of Technology and Science Pilani xPilani Campus Vidya Vihar Rajasthan Pilani333031 India Department of Computer Science and Information Systems Birla Institute of Technology and Science Pilani Pilani Campus Vidya Vihar Rajasthan Pilani333031 India

Transformers have emerged as a groundbreaking architecture in the field of computer vision, offering a compelling alternative to traditional convolutional neural networks (CNNs) by enabling the modeling of long-range dependencies and global context through self-attention mechanisms. Originally developed for natural language processing, transformers have now been successfully adapted for a wide range of vision tasks, leading to significant improvements in performance and generalization. This survey provides a comprehensive overview of the fundamental principles of transformer architectures, highlighting the core mechanisms such as self-attention, multi-head attention, and positional encoding that distinguish them from CNNs. We delve into the theoretical adaptations required to apply transformers to visual data, including image tokenization and the integration of positional embeddings. A detailed analysis of key transformer-based vision architectures such as ViT, DeiT, Swin Transformer, PVT, Twins, and CrossViT are presented, alongside their practical applications in image classification, object detection, video understanding, medical imaging, and cross-modal tasks. The paper further compares the performance of vision transformers with CNNs, examining their respective strengths, limitations, and the emergence of hybrid models. Finally, current challenges in deploying ViTs, such as computational cost, data efficiency, and interpretability, and explore recent advancements and future research directions including efficient architectures, self-supervised learning, and multimodal integration are discussed. © 2013 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

VTensor:Using Virtual Tensors to Build a Layout-Oblivious AI Programming Framework

引用

Journal of computer science & technology 2023年第5期38卷 1074-1097页

作者：俞峰赵家程崔慧敏冯晓兵薛京灵 Institute of Computing Technology Chinese Academy of SciencesBeijing 100190China School of Computer Science and Technology University of Chinese Academy of SciencesBeijing 100080China School of Computer Science and Engineering University of New South WalesSydney 1466Australia

Tensors are a popular programming interface for developing artificial intelligence(AI)*** refers to the order of placing tensor data in the memory and will affect performance by affecting data locality;therefore the deep neural network library has a convention on the *** AI applications can use arbitrary layouts,and existing AI systems do not provide programming abstractions to shield the layout conventions of libraries,operator developers need to write a lot of layout-related code,which reduces the efficiency of integrating new libraries or developing new ***,the developer assigns the layout conversion operation to the internal operator to deal with the uncertainty of the input layout,thus losing the opportunity for layout *** on the idea of polymorphism,we propose a layout-agnostic virtual tensor programming interface,namely the VTensor framework,which enables developers to write new operators without caring about the underlying physical layout of *** addition,the VTensor framework performs global layout inference at runtime to transparently resolve the required layout of virtual tensors,and runtime layout-oriented optimizations to globally minimize the number of layout transformation *** results demonstrate that with VTensor,developers can avoid writing layout-dependent *** with TensorFlow,for the 16 operations used in 12 popular networks,VTensor can reduce the lines of code(LOC)of writing a new operation by 47.82%on average,and improve the overall performance by 18.65%on average.

关键词： artificial intelligence(AI)programming layout-oblivious tensor processing

来源：评论

学校读者我要写书评

暂无评论

FERMixNet: An Occlusion Robust Facial Expression Recognition Model with Facial Mixing Augmentation and Mid-Level Representation Learning

引用

IEEE Transactions on Affective Computing 2024年第2期16卷 639-654页

作者： Huang, Yansong Peng, Junjie Zhang, Wenqiang Zhao, Tong Chen, Gan Tan, Shuhua Yi, Fen Wang, Lu Shanghai University School of Computer Engineering and Science Shanghai China Shanghai University School of Computer Engineering and Science The Shanghai Institute for Advanced Communication and Data Science Shanghai China Fudan University Academy for Engineering and Technology School of Computer Science and Technology Shanghai China YTO Express Co. Ltd. The National Logistics Engineering Laboratory Shanghai China

Facial expressions can provide a better understanding of people's mental status and attitudes towards specific things. However, facial occlusion in real world is an unfavorable phenomenon that greatly affects the performance of facial expression recognition models. Recent works addressing the occlusion problem have primarily relied on attention mechanisms or occlusion discarding methods that focus on non-occluded regions of the face. However, these methods have not achieved a good balance between occlusion robustness and model efficiency. In this paper, we propose a simple and efficient model, called FERMixNet, for occluded facial expression recognition. The model incorporates a novel facial mixing augmentation strategy (FERMix) that generates new training samples by simulating real-world facial occlusion and preserving high expression-related semantic information. By co-training the original and newly generated samples, the model's occlusion robustness is improved without increasing its complexity during inference. Additionally, to further enhance the model's occlusion robustness, we include mid-level representation learning in the network to learn the discriminative non-occluded local features of the samples with low computational cost. Extensive experiments on four public facial occlusion datasets: Occlusion-RAF-DB, Occlusion-FERPlus and FED-RO show that the proposed model achieves state-of-the-art results which demonstrates the good robustness of our method for occluded facial expression recognition. Meanwhile, the proposed model also achieves state-of-the-art results on the in-the-wild facial expression datasets RAF-DB, AffectNet-8, and AffectNet-7. It proves that the proposed model has good application prospects in real world. © 2010-2012 IEEE.

关键词： Face recognition

来源：评论

学校读者我要写书评

暂无评论

Target extraction through strong scattering disturbance using characteristic-enhanced pseudo-thermal ghost imaging

引用

Chinese Optics Letters 2024年第12期22卷 38-44页

作者： Xuanpengfan Zou Xianwei Huang Wei Tan Liyu Zhou Xiaohui Zhu Qin Fu Xiaoqian Liang Suqin Nan Yanfeng Bai Xiquan Fu College of Computer Science and Electronic Engineering Hunan UniversityChangsha 410082China Hunan Police Academy Changsha 410138China Hunan Provincial Key Laboratory of Network Investigational Technology Hunan Police AcademyChangsha 410138China School of Computer Science Hunan University of Technology and BusinessChangsha 410205China

It is difficult to extract targets under strong environmental disturbance in *** imaging(GI)is an innovative antiinterference imaging *** this paper,we propose a scheme for target extraction based on characteristicenhanced pseudo-thermal *** traditional GI which relies on training the detected signals or imaging results,our scheme trains the illuminating light fields using a deep learning network to enhance the target’s characteristic *** simulation and experimental results prove that our imaging scheme is sufficient to perform single-and multiple-target extraction at low *** addition,the effect of a strong scattering environment is discussed,and the results show that the scattering disturbance hardly affects the target extraction *** proposed scheme presents the potential application in target extraction through scattering media.

关键词： target extraction ghost imaging characteristic enhancement strong scattering environment

来源：评论

学校读者我要写书评

暂无评论

circ2CBA: prediction of circRNA-RBP binding sites combining deep learning and attention mechanism

引用

Frontiers of computer science 2023年第5期17卷 217-225页

作者： Yajing GUO Xiujuan LEI Lian LIU Yi PAN School of Computer Science Shaanxi Normal UniversityXi’an 710119China Faculty of Computer Science and Control Engineering Shenzhen Institute of Advanced TechnologyChinese Academy of SciencesShenzhen 518055China

Circular RNAs(circRNAs)are RNAs with closed circular structure involved in many biological processes by key interactions with RNA binding proteins(RBPs).Existing methods for predicting these interactions have limitations in feature *** view of this,we propose a method named circ2CBA,which uses only sequence information of circRNAs to predict circRNA-RBP binding *** have constructed a data set which includes eight ***,circ2CBA encodes circRNA sequences using the one-hot ***,a two-layer convolutional neural network(CNN)is used to initially extract the *** CNN,circ2CBA uses a layer of bidirectional long and short-term memory network(BiLSTM)and the self-attention mechanism to learn the *** AUC value of circ2CBA reaches *** of circ2CBA with other three methods on our data set and an ablation experiment confirm that circ2CBA is an effective method to predict the binding sites between circRNAs and RBPs.

关键词： circRNAs RBPs CNN BiLSTM self-attention mechanism

来源：评论

学校读者我要写书评

暂无评论

Price Prediction of Resale Vehicle Through Supervised Learning Approaches 4

Price Prediction of Resale Vehicle Through Supervised Learni...

引用

4th International Conference on computer, Communication, Control and Information technology, C3IT 2024

作者： Paul, Ankit Guchait, Chiranjit Bhukta, Priyanath Computer Science Engineering Academy of Technology Adisaptagram India Mechanical Engineering Academy of Technology Adisaptagram India

ISBN: (纸本)9798350389128

The world is changing day by day and the expectations of every individual are also becoming high. Out of all the expectations, one of them is to buy a car. But all are not able to always buy a new car, so they will prefer a used one. In this study, we have conducted an analysis to help individuals make informed decisions about used car prices. By using supervised learning algorithms like multiple linear regressions and random forest regression along with Python, we have developed a model which can predict the price of resale vehicles. The dataset used here for prediction has been taken from *** and the final accuracy (R2) of multiple linear regression is 0.93464 and for random forest regression is 0. 93907. © 2024 IEEE.

关键词： Multiple linear regression

来源：评论

学校读者我要写书评

暂无评论

Align is not Enough: Multimodal Universal Jailbreak Attack against Multimodal Large Language Models

引用

IEEE Transactions on Circuits and Systems for Video technology 2025年第6期35卷 5475-5488页

作者： Wang, Youze Hu, Wenbo Dong, Yinpeng Liu, Jing Zhang, Hanwang Hong, Richang Hefei University of Technology School of Computer Science and Information Engineering Hefei230009 China Tsinghua University Beijing100084 China Chinese academy of science Institute of automation Beijing100190 China Nanyang Technological University School of Computer Science and Engineering 639798 Singapore

Large Language Models (LLMs) have evolved into Multimodal Large Language Models (MLLMs), significantly enhancing their capabilities by integrating visual information and other types, thus aligning more closely with the nature of human intelligence, which processes a variety of data forms beyond just text. Despite advancements, the undesirable generation of these models remains a critical concern, particularly due to vulnerabilities exposed by text-based jailbreak attacks, which have represented a significant threat by challenging existing safety protocols. Motivated by the unique security risks posed by the integration of new and old modalities for MLLMs, we propose a unified multimodal universal jailbreak attack framework that leverages iterative image-text interactions and transfer-based strategy to generate a universal adversarial suffix and image. Our work not only highlights the interaction of image-text modalities can be used as a critical vulnerability but also validates that multimodal universal jailbreak attacks can bring higher-quality undesirable generations across different MLLMs. We evaluate the undesirable context generation of MLLMs like LLaVA, Yi-VL, MiniGPT4, MiniGPT-v2, and InstructBLIP, and reveal significant multimodal safety alignment issues, highlighting the inadequacy of current safety mechanisms against sophisticated multimodal attacks. This study underscores the urgent need for robust safety measures in MLLMs, advocating for a comprehensive review and enhancement of security protocols to mitigate potential risks associated with multimodal capabilities. © 1991-2012 IEEE.

关键词： Human form models

来源：评论

学校读者我要写书评

暂无评论

A Survey of Edge Caching:Key Issues and Challenges

引用

Tsinghua science and technology 2024年第3期29卷 818-842页

作者： Hanwen Li Mingtao Sun Fan Xia Xiaolong Xu Muhammad Bilal School of Computer Science Nanjing University of Information Science and TechnologyNanjing 210044China Shandong Provincial University Laboratory for Protected Horticulture Weifang University of Science and TechnologyWeifang 262700China Reading Academy Nanjing University of Information Science and TechnologyNanjing 210044China School of Software Nanjing University of Information Science and TechnologyNanjing 210044China Department of Computer Engineering Hankuk University of Foreign StudiesYongin-si 17035Republic of Korea

With the rapid development of mobile communication technology and intelligent applications,the quantity of mobile devices and data traffic in networks have been growing exponentially,which poses a great burden to networks and brings huge challenge to servicing user *** caching,which utilizes the storage and computation resources of the edge to bring resources closer to end users,is a promising way to relieve network burden and enhance user *** this paper,we aim to survey the edge caching techniques from a comprehensive and systematic *** first present an overview of edge caching,summarizing the three key issues regarding edge caching,i.e.,where,what,and how to cache,and then introducing several significant caching *** then carry out a detailed and in-depth elaboration on these three issues,which correspond to caching locations,caching objects,and caching strategies,*** particular,we innovate on the issue“what to cache”,interpreting it as the classification of the“caching objects”,which can be further classified into content cache,data cache,and service ***,we discuss several open issues and challenges of edge caching to inspire future investigations in this research area.

关键词： edge caching edge computing caching location caching object caching strategy 5G networkarchitecture Internet of Things(IoT)

来源：评论

学校读者我要写书评

暂无评论

Design of high-speed, low-power non-volatile master slave flip flop (NVMSFF) for memory registers designs

引用

Applied Nanoscience (Switzerland) 2023年第8期13卷 5369-5378页

作者： Muthappa, K.A. Nisha, A. Sahaya Anselin Shastri, Rajveer Avasthi, Vinay Reddy, Pundru Chandra Shaker Department of ECE Sathyabama Institute of Science and Technology Chennai India Department of E and TC Engineering Vidya Pratishthan’s Kamalnayan Bajaj Institute of Engineering and Technology Baramati413133 India School of Computer Science University of Petroleum and Energy Studies Dehradun Dehradun India Computer Science and Engineering Geethanjali College of Engineering and Technology Telangana Hyderabad India

High-speed and low-power area-efficient memory solutions are in high demand in today’s smart and internet environment. To program and store data and its associated information, it has grown as a fantastic concept to replace charge in semiconductor devices with electron spin. By further eliminating power dissipation, this leads to the development of "green electronics," a field of technology that is both energy and chip space efficient. A sort of spintronic device created for potential energy-efficient applications is the magnetic tunnel junction (MTJ) device controlled by spin orbit torques (SOTs). A three-terminal MTJ device with perpendicular to the plane magnetization may operate more quickly and efficiently thanks to the available spin orbit torques (PMTJ). The design of a novel modified non-volatile master slave flip flop (NVMSFF) for register memory array architectures is presented in this study. It is low-power, high-speed, and area-efficient. The NVMSFF design uses a hybrid approach that combines CMOS 45 nm technology with a Verilog-A Model-based SOT-PMTJ circuit that uses spintronics. Additionally, the planned NVMSFF and current NVSFF were used in the design of the 4 × 4 and 8 × 8 memory arrays. The suggested updated NVMSFF was shown to have a lower latency, use less space, and dissipate less power. Additionally, it was discovered that Designed Memory arrays were space and energy efficient. © 2023, King Abdulaziz City for science and technology.

关键词： CMOS integrated circuits

来源：评论

学校读者我要写书评

暂无评论

An autoencoder based unsupervised clustering approach to analyze the effect of E-learning on the mental health of Indian students during the Covid-19 pandemic

引用

Multimedia Tools and Applications 2024年 1-21页

作者： Banerjee, Pritha Jana, Chandan Saha, Jayita Chowdhury, Chandreyee Computer Application and Science Institute of Engineering and Management West Bengal Kolkata700091 India Computer Science and Engineering Jadavpur University West Bengal Kolkata700032 India Computer Science and Engineering Manipal Institute of Technology Bengaluru Manipal Academy of Higher Education Manipal Karnataka Bengaluru560064 India

Due to the Covid-19 pandemic, the education system in India has changed to remote that is, online study mode. Though there are works on the effect of teaching learning on Indian students, the effect of online mode and associated mental state, particularly when the entire country is going through a crisis could not be found in the literature. Our goal is to analyze data and find some pattern through which we can understand the effectiveness of the online study and also try to figure out the stress level. The dataset we collected from 500 undergraduate college students during April-May, 2021 is in questionnaire format. Our contribution in this paper are - (i) publishing a dataset of student feedbacks, and (ii) designing a data processing pipeline involving autoencoders followed by clustering approach. The dataset is in text format so for our analysis we have converted the dataset into a numerical format using the concept of a binary bag of words. Dimensionality reduction is applied through autoencoder for an effective latent space representation. Finally, for finding patterns out of this dimensionally reduced feature space, we have applied unsupervised learning algorithms - kMeans and DBSCAN. A thorough analysis of the clustering process reveals that the absence of social communication in purely online education provokes isolation irrespective of the urban or rural background of the students. However, it could supplement offline classes as a substantial number of students welcomed the concept of online learning as reported in the data. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Students

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：