检索结果-内蒙古大学图书馆

Explaining Internal Representations in Deep Networks: Adversarial Vulnerability of Image Classifiers and Learning Sequential Tasks with Sparse Reward

Explaining Internal Representations in Deep Networks: Advers...

引用

International symposium on Applied Machine Intelligence and Informatics (SAMI)

作者： Igor Farkaš Department of Applied Informatics Physics and Informatics Centre for Cognitive Science Faculty of Mathematics Comenius University Bratislava Slovak Republic

ISBN: (数字)9798350379365

ISBN: (纸本)9798350379372

Modern artificial intelligence based on deep neural networks has demonstrated in the past decade great achievements in concrete tasks, sometimes even surpassing human performance. On the other hand, there exist fundamental problems in these models, be it image classification or natural language tasks. Since deep networks are inherently black boxes, it is important to design and apply techniques that help shed light on the functioning of the trained models. In the talk, we will discuss two domains. First, in the context of image classification we will illustrate the effect of adversarial examples that can easily fool trained models, hence revealing their lack of robustness. In the second part, we will deal with sequential tasks, such as computer games, with an extremely sparse reward. Introducing the concept of intrinsic motivation, we will describe the neural networks based model that can successfully use reinforcement learning and self-supervised knowledge distillation to solve these tasks thanks to optimized organization of its internal representations. Finally, we briefly mention our current work related to cognitive robotics.

关键词： Knowledge engineering Video games Computational modeling Natural languages Reinforcement learning Organizations Robustness Informatics Machine intelligence Image classification

来源：评论

学校读者我要写书评

暂无评论

Ordinal focal loss: A relative fashionability ranking learning method

引用

Information Processing and Management 2025年第5期62卷

作者： Ma, Ling Zou, Chuhang Guo, Ziyi Li, Tao Liu, Zheng Zou, Fengyuan Key Laboratory of Silk Culture Inheriting and Products Design Digital Technology Ministry of Culture and Tourism Zhejiang Sci-Tech University Linping Campus Zhejiang Hangzhou310018 China Amazon 410 Terry Ave N SeattleWA98109 United States School of Fashion Design & Engineering Zhejiang Sci-Tech University Linping Campus Zhejiang Hangzhou310018 China Clothing Engineering Research Center of Zhejiang Province Zhejiang Sci-Tech University Linping Campus Zhejiang Hangzhou310018 China Zhejiang International Institute of Fashion Technology Zhejiang Sci-Tech University Linping Campus Zhejiang Hangzhou310018 China

Existing fashion recommendations often rely on natural language processing or content-based image retrieval, overlooking direct aesthetic assessments of fashion images. Given the subjectivity and complexity of this task, we propose treating fashionability as a relative attribute to rank paired clothing images. To address this ranking challenge, we propose Ordinal Focal Loss, which transforms the pairwise ranking problem into a multi-classification task, leveraging ordinal attributes to improve classification boundaries. Furthermore, in terms of fashion feature representation, we propose modeling not just individual items but also their combined effect as an outfit, providing a more holistic and nuanced fashion representation. We introduce the Fashionability3k dataset, comprising 3k image pairs (2398 ordered and 601 similar pairs) with objective relative fashion labels. Experiments on three datasets—our Fashionability3k and two public datasets—show that our method outperforms the baseline by nearly 1 % in ranking accuracy. In the user study, it achieved a 0.72 consistency with human subjective perception. Moreover, combining local and global visual features leads to additional performance gains, with an average improvement of 2.78 % in ordered pairs and 1.09 % in similar pairs. This is the first study to treat fashionability as an objective attribute for comparative analysis, validated through extensive experiments. © 2025 Elsevier Ltd

关键词： Content based retrieval

来源：评论

学校读者我要写书评

暂无评论

Re-HOLD: Video Hand Object Interaction Reenactment via adaptive Layout-instructed Diffusion Model

arXiv

引用

arXiv 2025年

作者： Fan, Yingying Yang, Quanwei Wang, Kaisiyuan Zhou, Hang Li, Yingying Feng, Haocheng Ding, Errui Wu, Yu Wang, Jingdong School of Computer Science Wuhan University China University of Science and Technology of China China Baidu Inc. China

Current digital human studies focusing on lip-syncing and body movement are no longer sufficient to meet the growing industrial demand, while human video generation techniques that support interacting with real-world environments (e.g., objects) have not been well investigated. Despite human hand synthesis already being an intricate problem, generating objects in contact with hands and their interactions presents an even more challenging task, especially when the objects exhibit obvious variations in size and shape. To tackle these issues, we present a novel video reenactment framework focusing on human-Object Interaction (HOI) via an adaptive Layout-instructed Diffusion model (Re-HOLD). Our key insight is to employ specialized layout representation for hands and objects, respectively. Such representations enable effective disentanglement of hand modeling and object adaptation to diverse motion sequences. To further improve the quality of the HOI generation, we design an interactive textural enhancement module for both hands and objects by introducing two independent memory banks. We also propose a layout adjustment strategy for the cross-object reenactment scenario to adaptively adjust unreasonable layouts caused by diverse object sizes during inference. Comprehensive qualitative and quantitative evaluations demonstrate that our proposed framework significantly outperforms existing methods. Project page: https://***/Re-HOLD. Copyright © 2025, The Authors. All rights reserved.

关键词： human engineering

来源：评论

学校读者我要写书评

暂无评论

服饰博物馆的数字化发展历程和研究现状

引用

服装学报 2025年第2期10卷 101-109页

作者：王旭万腾淑田霞徐忠林张育玮河南工程学院服装学院河南郑州451191 中原工学院纺织服装产业研究院河南郑州450007 天津工业大学纺织科学与工程学院天津300387

为加快服饰博物馆的数字化建设进程,满足多元化观览需求,对国内外服饰博物馆的发展历程、技术手段、存在问题等方面的文献进行梳理和分析,探讨服饰博物馆数字化建设的有效途径。研究表明,通过数字信息采集与建模、数字化展示平台建设及... 详细信息

为加快服饰博物馆的数字化建设进程,满足多元化观览需求,对国内外服饰博物馆的发展历程、技术手段、存在问题等方面的文献进行梳理和分析,探讨服饰博物馆数字化建设的有效途径。研究表明,通过数字信息采集与建模、数字化展示平台建设及用户交互体验设计等方式,可实现传统服饰文化的数字化展示和传播,打造人性化、沉浸式的博物馆展览模式。服饰博物馆在数字服饰体系构建、服务平台共享、用户体验设计等方面还存在一定的问题,今后应着重从专业性、开放性、用户友好性等方向入手,开展系统的理论和技术研究工作,推动数智化建设,实现传统服饰文化展示与现代技术的深度融合,有效助力中华优秀传统服饰文化的传承与发展。

关键词：传统服饰服饰博物馆数字化技术人机交互技术文化传承

来源：评论

学校读者我要写书评

暂无评论

HeXA: Haptic-enhanced eXtended reality framework for material-informed Architectural design

引用

Journal of Building engineering 2025年 108卷

作者： Senk, Valentin Ghazanfari, Mohammad Rasoulzadeh, Shervin Vasylevska, Khrystyna Kovacs, Balint Istvan Mortezapoor, Soroosh Vonach, Emanuel Kovacic, Iva Füssl, Josef Kaufmann, Hannes Königsberger, Markus Institute of Mechanics of Materials and Structures Vienna University of Technology Karlsplatz 13/202 Vienna1040 Austria Institute of Visual Computing and Human-Centered Technology Vienna University of Technology Favoritenstrasse 9-11 Vienna1040 Austria Institute of Building and Industrial Construction Vienna University of Technology Karlsplatz 13/210 Vienna1040 Austria Institute of Architectural Sciences Vienna University of Technology Treitlstrasse 3 Vienna1040 Austria

Traditional architectural design processes are caught in sequential and isolated planning workflows, starting with pen-and-paper sketching, digitizing, visualizing for stakeholders, and refining. Only after the basic design has been established, structural engineers get involved and modify the structure, such that the design withstands all external loads. This segregated approach is driven by knowledge barriers associated with structural analysis and by lengthy feedback loops regarding structural design decisions involving finite element simulations. This paper presents research on a novel collaborative and digital design framework, labeled as HeXA, that lowers this knowledge barrier and enables all stakeholders to actively participate in structural design decisions. HeXA couples architecture, structural and material mechanics, computer science and graphics, as well as robotics in six modules: (1) a sketching interface for architectural design, (2) a geometric modeling module that translates the sketch to finite element meshes, (3) a material modeling module that predicts the mechanical properties of sustainable bio-composites, (4) a structural analysis module that assesses the mechanical performance using finite element simulations, and (5) an extended reality environment's immersive visualization that is enhanced by (6) structural haptic feedback provided by a collaborative robot. This allows users to touch and push the envisioned structure in a virtual environment and experience realistic structural deformations provided by the robot and its arm in real-time while resulting stress fields are mapped onto the virtual structure. This way, architectural and structural design processes – particularly at the crucial early design stage – become collaborative, interactive, accessible, and responsive as demonstrated in the paper based on two use cases in theater stage design and building architecture. © 2025 The Authors

关键词： Architectural design

来源：评论

学校读者我要写书评

暂无评论

EMR usability and patient safety: a national survey of physicians

引用

npj digital Medicine 2025年第1期8卷 1-8页

作者： Schwappach, David Hautz, Wolf Krummrey, Gert Pfeiffer, Yvonne Ratwani, Raj M. University of Bern Bern Switzerland Department of Emergency Medicine Inselspital University Hospital University of Bern Bern Switzerland Biel/Bienne Switzerland harmfree healthcare Wädenswil Switzerland Faculty of Medicine University of Basel Basel Switzerland MedStar Health National Center for Human Factors in Healthcare WashingtonDC United States Georgetown University School of Medicine WashingtonDC United States

Despite widespread adoption of electronic medical records (EMRs), concerns persist regarding their usability and implications for patient safety. This national cross-sectional survey assessed physicians’ perceptions of EMR usability across safety-relevant domains. Among 1933 respondents from diverse care settings, 56% reported that their EMR did not enhance patient safety, and 50% perceived their system as inefficient. Usability ratings averaged 52% of the maximum score. Statistically significant differences were observed between EMRs in outpatient (η² = 0.13) and hospital (η² = 0.37) settings. Multilevel modeling attributed 38% of the variance in usability ratings to differences between EMRs, 51% to hospital-level variation within EMRs, and 11% to physician-level differences. Canonical discriminant analysis identified key differentiating usability features, including system response times, excessive alerts, prevention of data entry errors, and support for collaboration. These findings underscore substantial limitations in current EMR systems and reinforce the value of comparative usability assessments to inform targeted improvements in digital health infrastructure. © The Author(s) 2025.

关键词： Biomedical engineering

来源：评论

学校读者我要写书评

暂无评论

C2OPU: Hybrid Compute-in-Memory and Coarse-Grained Reconfigurable Architecture for Overlay Processing of Transformers

C2OPU: Hybrid Compute-in-Memory and Coarse-Grained Reconfigu...

引用

Annual IEEE symposium on Field-Programmable Custom Computing Machines (FCCM)

作者： Siyuan Miao Lingkang Zhu Chen Wu Shaoqiang Lu Jinming Lyu Lei He University of California Los Angeles United States University of Nottingham Ningbo China Ningbo Institute of Digital Twin Eastern Institute of Technology Ningbo China Engineering Research Center of Chiplet Design and Manufacturing of Zhejiang Province Ningbo Zhejiang China

ISBN: (数字)9798331502812

ISBN: (纸本)9798331502829

Transformer-based models have shown huge success in natural language processing (NLP) with increasing model size and attention mechanism. However, this makes Von Neumann architecture based accelerators memory-bound such that the accelerators cannot leverage all the advantages of Transformer-based models. Although computing-in-memory (CIM) processors have emerged to tackle this problem through in-situ computing, the mismatch of computing patterns and low computing precision of CIM make it still challenging to accelerate Transformers. In this paper, we propose C2OPU, a hybrid dual-core processor to accelerate Transformers with hardware and software co-optimization. The dual-core architecture uses CIM arrays to accelerate weight-stationary vector-matrix multiplications, which accounts for the main computation complexity of Transformers. Meanwhile, a coarse-grained reconfigurable architecture (CGRA) is used to address the issues of computing pattern mismatch and low precision of the CIM. In addition, we propose an accuracy-bound workload allocation strategy, which considers non-ideal characteristics in analog computing, to balance throughput and accuracy. Furthermore, C2OPU provides a compiler to automatically determine optimal system configurations when Transformer model changes. Experimental results show that C2OPU achieves an average speedup of 145.41×, 4.73×, 4.70x and 3.85×, and 1.37x compared to CPU, GPU, Science23, Nature23, and VLSI24, respectively, on ten different Transformer models.

关键词： Computational modeling Memory architecture In-memory computing Transformers Throughput Common Information Model (computing) Natural language processing Software Reconfigurable architectures Resource management

来源：评论

学校读者我要写书评

暂无评论

Superior Flexible Tactile Sensor with AI-based Unit Microstructure design for human and Robot Parameters Monitoring Application

引用

IEEE Sensors Journal 2025年

作者： Yao, Pei Liu, Zhuo Xiao, Xuan Duo, Yongchao Dai, Guang Wang, Liang Xu, Hongchen Tiangong University Tianjin Key Laboratory of Autonomous Intelligence Technology and Systems School of Computer Science and Technology Tianjin300387 China Tiangong University School of Textiles Science and Engineering Tianjin300387 China Tianjin University of Technology School of Science Key Laboratory of Quantum Optics and Intelligent Photonics School of Science Tianjin300384 China Shanxi Xi'an710049 China Xi'an Jiaotong University School of Instrument Science and Technology Shanxi Xi'an710049 China

Due to the development of wearable monitoring systems and the gradually expanding application demand, research interest has turned to free-form microstructures design capable of realizing superior and stable piezoresistive responses. The deep learning methods can accelerate and improve microstructure design with data-driven precision and increased efficiency. In this study, the concept of programmable microstructure is introduced to digitalize microstructure unit, convolutional neural network (CNN) model and Tabu-Search algorithm are respectively used to predict and search iteratively novel stress-voltage (S-V) responses from the finite element modeling (FEM) training set. Therefore, efficient exploration of parameter space and faster generating of novel microstructure designs can be achieved by the proposed algorithm framework, and the optimized microstructure has also been experimentally validated. Moreover, the working mechanism of the sensing medium layer is explicated by the resistance evolution of Ti2CTx-MXene atomic layers under the compressive strain via density functional theory calculations. The proposed AI-based tactile sensor exhibits high sensitivity, large linearity, fast response, and excellent cycling stability, it leverages strong capabilities to detect human being's subtle activities and terrain perception of snake robot. This study could achieve a higher degree in human physiological monitoring and robot tactile construction, and gain a new insight for digital microstructure in sensor design. © 2001-2012 IEEE.

关键词： Programmable robots

来源：评论

学校读者我要写书评

暂无评论

Bayesian neural networks for probabilistic modeling of thermal dynamics in multiscale tissue engineering scaffolds

引用

Journal of Thermal Biology 2025年 130卷 104134页

作者： Naga Ramesh, Janjhyam Venkata Sonker, Abhilash Indumathi, G. Balakrishnan, D. Nimma, Divya Karthik, Jayavarapu Department of CSE Graphic Era Hill University Dehradun 248002 India Department of Information Technology MITS Madhya Pradesh Gwalior India Department of CSE SRM Institute of Science and Technology Ramapuram Tamil Nadu Chennai India Department of Computer Science and Engineering Kalasalingam Academy of Research and Education Krishnankoil Tamil Nadu 626126 India Computer Science & Digital Technologies Grambling State University 403 Main Street Grambling 71245 LA United States Department of Computer Science and Engineering Koneru Lakshmaiah Education Foundation Andhra Pradesh Vaddeswaram 522502 India Department of CSE Graphic Era Deemed To Be University Uttarakhand Dehradun 248002 India

Multiscale tissue engineering integrated with thermal dynamics exhibits a critical role as scaffolds in maintaining the structural integrity and functionality in fabrication applications. Thermal dynamics influence the properties such as thermal conductivity, the capacity of heat, and thermal expansion of materials to maintain the scaffold stability in various physiological conditions. However, the scaffold's heat distribution in a uniform manner is varied due to variations in pore size and geometrics. Additionally, variation in scales affects the thermal gradient impacts on the cell growth and integrity. This paper proposes the 3D Scaffolds Probabilistic Weighted Bayesian Neural Network (3D-SP-WBNN) for tissue engineering with a multiscale scaffold model. It uses the 3D scaffolds for the multiscale design estimation. The weighted Probabilistic Bayesian Neural Network model is employed for estimating features in tissues and evaluating the cell growth and proliferation. Thermal gradients measured were in between 1 °C and 4 °C for human bone, skin, and cartilage tissues. For low and moderate temperatures of 1 °C and 2 °C, the cell proliferation rates in cartilage tissues were 15–20 % per day. The scaffold design uses the Hybrid Hydrogel/PCL composite to achieve a higher proliferation rate of 25–35 % per day. The estimated forecasting achieves an accuracy range of 82–96 % for different cell densities and thermal conditions. © 2025 Elsevier Ltd

关键词： Bayesian neural network Cell growth Multiscale tissue Probabilistic network Thermal expansion Tissue engineering

来源：评论

学校读者我要写书评

暂无评论

Structure-Preserving 3D Reconstruction Algorithm for Traditional Buildings Based on Hybrid Imaging Techniques

Structure-Preserving 3D Reconstruction Algorithm for Traditi...

引用

International Conference on Inventive Computation Technologies (ICICT)

作者： Guangpei Yang Yue Zhong School of Architectural Engineering and Art Guangzhou Nanyang Polytechnic Guangzhou Guangdong China School of Information Science and Technology Guangdong Construction Polytechnic Guangzhou Guangdong China

ISBN: (数字)9798331512248

ISBN: (纸本)9798331512255

This study proposes a novel traditional building structure-preserving 3D reconstruction algorithm based on hybrid imaging technology. The main application scenario of this algorithm is to provide a digital solution for building protection for the deformation and degradation of the traditional buildings caused by natural or human factors. This study first analyzes RGB imaging, thermal imaging, and 3D point cloud as three mainstream imaging technologies, and conducts a preliminary analysis of different modes of image analysis. At the algorithm design level, this study first uses the cross-modal registration algorithm, SIFT feature extraction and neural network density optimization method for comprehensive modeling. This operation achieves accurate alignment of RGB images, thermal imaging images and 3D point cloud data. Then, in the image detail optimization stage. This study uses high-frequency detail enhancement technology based on wavelet transform to enhance details. This operation enhances the texture and edge information of the image. Finally, combined with the partial completion algorithm of the image style transfer. Local style transfer is performed on the target area under the style prior background, thereby completing the restoration and reconstruction of the architectural image. Experimental results show that the proposed algorithm is superior to the traditional algorithm in image restoration quality, and obtains good performance in specific indicators, such as PSNR value of 32.0 dB, SSIM value of 0.85, and LPIPS value of 0.08.

关键词： Point cloud compression Wavelet transforms Three-dimensional displays Accuracy Image edge detection Buildings Imaging Thermal conductivity Image restoration Image reconstruction

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：