检索结果-内蒙古大学图书馆

arXiv 2025年

作者： Jo, Sanghyun Lee, Ziseok Lee, Wooyeol Kim, Kyungsu OGQ Seoul Korea Republic of Department of Biomedical Sciences Seoul National University Seoul Korea Republic of Department of Computer Science and Engineering Seoul National University Seoul Korea Republic of School of Transdisciplinary Innovations Interdisciplinary Program in Artificial Intelligence Seoul National University Seoul Korea Republic of

Achieving precise panoptic segmentation heavily relies on pixel-wise instance annotations, but obtaining such datasets is costly. Unsupervised instance segmentation (UIS) eliminates annotation requirements but struggles with adjacent instance merging and single-instance fragmentation, largely due to the limitations of DINO-based backbones, which lack strong instance separation cues. Weakly-supervised panoptic segmentation (WPS) reduces annotation costs by using sparse labels (e.g., points, boxes), yet these annotations remain expensive and introduce human bias and boundary errors, limiting performance. To address these challenges, we propose DiffEGG (Diffusion-Driven EdGe Generation), a fully annotation-free method that extracts instance-aware features from pretrained diffusion models to generate precise instance edge maps. DiffEGG fundamentally differs from prior UIS and WPS approaches by leveraging pretrained diffusion models as a backbone. Unlike DINO-based UIS methods, diffusion models inherently capture fine-grained, instance-aware features, enabling more precise boundary delineation. DiffEGG further enhances edge connectivity through self-supervised fine-tuning, progressively improving instance separation. For WPS, DiffEGG completely eliminates annotation costs and human bias by operating without any form of manual supervision, addressing the key limitations of prior best methods. Additionally, we introduce RIP, a post-processing technique that fuses DiffEGG’s edge maps with segmentation masks in a task-agnostic manner. This broad applicability allows DiffEGG to be seamlessly integrated into various segmentation frameworks. When applied to UIS, DiffEGG and RIP achieve an average +4.4 AP improvement over prior best UIS methods. When combined with weakly-supervised semantic segmentation (WSS), DiffEGG enables WPS without instance annotations, outperforming prior best point-supervised WPS methods by +1.7 PQ. These results demonstrate that DiffEGG’s edge maps serv

关键词： Semantic Segmentation

来源：评论

学校读者我要写书评

暂无评论

MathSpeech: Leveraging Small LMs for Accurate Conversion in Mathematical Speech-to-Formula 39

MathSpeech: Leveraging Small LMs for Accurate Conversion in ...

引用

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： Hyeon, Sieun Jung, Kyudan Won, Jaehee Kim, Nam-Joon Ryu, Hyun Gon Lee, Hyuk-Jae Do, Jaeyoung Department of Electrical and Computer Engineering Seoul National University Korea Republic of Department of Mathematics Chung-Ang University Korea Republic of College of Liberal Studies Seoul National University Korea Republic of NVIDIA United States Interdisciplinary Program in Artificial Intelligence Seoul National University Korea Republic of

ISBN: (纸本)157735897X

In various academic and professional settings, such as mathematics lectures or research presentations, it is often necessary to convey mathematical expressions orally. However, reading mathematical expressions aloud without accompanying visuals can significantly hinder comprehension, especially for those who are hearing-impaired or rely on subtitles due to language barriers. For instance, when a presenter reads Euler's Formula, current Automatic Speech Recognition (ASR) models often produce a verbose and error-prone textual description (e.g., e to the power of i x equals cosine of x plus i side of x), instead of the concise LATEX format (i.e., eix = cos(x) + isin(x)), which hampers clear understanding and communication. To address this issue, we introduce MathSpeech, a novel pipeline that integrates ASR models with small Language Models (sLMs) to correct errors in mathematical expressions and accurately convert spoken expressions into structured LATEX representations. Evaluated on a new dataset derived from lecture recordings, MathSpeech demonstrates LATEX generation capabilities comparable to leading commercial Large Language Models (LLMs), while leveraging fine-tuned small language models of only 120M parameters. Specifically, in terms of CER, BLEU, and ROUGE scores for LATEX translation, MathSpeech demonstrated significantly superior capabilities compared to GPT-4o. We observed a decrease in CER from 0.390 to 0.298, and higher ROUGE/BLEU scores compared to GPT-4o. Copyright © 2025, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A Portable Multiscopic Camera for Novel View and Time Synthesis in Dynamic Scenes

arXiv

引用

arXiv 2022年

作者： Zhang, Tianjia Lau, Yuen-Fui Chen, Qifeng The Hong Kong University of Science and Technology Hong Kong The Individualized Interdisciplinary Program The Robotics Institute Hong Kong The Department of Mathematics Hong Kong The Department of Computer Science and Engineering The Department of Electronic and Computer Engineering Hong Kong

We present a portable multiscopic camera system with a dedicated model for novel view and time synthesis in dynamic scenes. Our goal is to render high-quality images for a dynamic scene from any viewpoint at any time using our portable multiscopic camera. To achieve such novel view and time synthesis, we develop a physical multiscopic camera equipped with five cameras to train a neural radiance field (NeRF) in both time and spatial domains for dynamic scenes. Our model maps a 6D coordinate (3D spatial position, 1D temporal coordinate, and 2D viewing direction) to view-dependent and time-varying emitted radiance and volume density. Volume rendering is applied to render a photo-realistic image at a specified camera pose and time. To improve the robustness of our physical camera, we propose a camera parameter optimization module and a temporal frame interpolation module to promote information propagation across time. We conduct experiments on both real-world and synthetic datasets to evaluate our system, and the results show that our approach outperforms alternative solutions qualitatively and quantitatively. Our code and dataset are available at https://***/. Copyright © 2022, The Authors. All rights reserved.

关键词： Cameras

来源：评论

学校读者我要写书评

暂无评论

Dynamic characteristics of terahertz hot-electron graphene FET bolometers: effect of electron cooling in channel and at side contacts

arXiv

引用

arXiv 2024年

作者： Ryzhii, V. Tang, C. Otsuji, T. Ryzhii, M. Mitin, V. Shur, M.S. Research Institute of Electrical Communication Tohoku University Sendai980-8577 Japan Frontier Research Institute for Interdisciplinary Sciences Tohoku University Sendai980-8578 Japan Department of Computer Science and Engineering University of Aizu Aizu-Wakamatsu965-8580 Japan Department of Electrical Engineering University at Buffalo SUNY BuffaloNY14260 United States Department of Electrical Computer and Systems Engineering Rensselaer Polytechnic Institute TroyNY12180 United States

We analyze the operation of the hot-electron FET bolometers with the graphene channels (GCs) and the gate barrier layers (BLs). Such bolometers use the thermionic emission of the hot electrons heated by incident modulated THz radiation. The hot electron transfer from the GC into the metal gate. As the THz detectors, these bolometers can operate at room temperature. We show that the response and ultimate modulation frequency of the GC-FET bolometers are determined by the efficiency of the hot-electron energy transfer to the lattice and the GC side contacts due to the 2DEG lateral thermal conductance. The dependences of these mechanisms on the band structure and geometrical parameters open the way for the GC-FET bolometers optimization, in particular, for the enhancement of the maximum modulation frequency. © 2024, CC BY.

关键词： Graphene

来源：评论

学校读者我要写书评

暂无评论

Classifying Ore Deposits Based on La-Icp-Ms Geochemistry of Natural Gold

SSRN

引用

SSRN 2024年

作者： Verbel, Angel Augusto Schutesky, Maria Emilia Gregory, Daniel D. Verbel, Arturo University of Brasilia Institute of Geosciences Graduate Program in Geology DF Brasília70910-900 Brazil University of Toronto Department of Earth Sciences TorontoONM5S 3B1 Canada Universidad de los Andes Systems and Computing Engineering Department School of Engineering Bogotá111711 Colombia

Systematic sampling of gold grains and their characterization in abundance and morphology can be used as a vectoring tool for mineral deposits. However, the processes that lead to the formation of an ore deposit are multiscale and nonlinear. As a result, the elemental compositions recorded in gold grains during the mineralization process seem mostly irregular and unpredictable. Here, we took advantage of the capabilities of neural networks to approximate, a natural nonlinear function that better explains the chemical variability observed in gold allowing us to make predictions. this was achieved by using published data of trace elements in gold of four types of gold deposits (i.e., Orogenic, VMS, porphyry and epithermal) from 47 different localities, obtained by laser ablation-inductively coupled plasma-mass spectrometry (LA- ICP-MS). The model was optimized by training five different architectures on the most influential elements, determined by principal component analysis (PCA). As a result, it was found that two hidden layers with 10 neurons each, is the best architecture for using trace elements as predictors of the type of deposit that formed a natural gold grain. In order to encourage the use of the findings made in this paper, we introduce OreGenes, an app developed in Matlab2023b based on the best-obtained model. It allows any user who possesses compatible data, the ability to import and process it to obtain a prediction with an average accuracy level of 88.9% confidence to determine the mineral deposit type that an unknown gold grain came from which grants the user the ability to have a powerful tool for exploration or research. © 2024, The Authors. All rights reserved.

关键词： Deposits

来源：评论

学校读者我要写书评

暂无评论

Endogenous transformation of land transport in Europe for different climate targets

arXiv

引用

arXiv 2025年

作者： Kalweit, Sina Zeyen, Elisabeth Victoria, Marta Department of Mechanical and Production Engineering Katrinebjergvej 89 Aarhus8200 Denmark Novo Nordisk Foundation CO2 Research Center Gustav Wieds Vej 10 Aarhus8000 Denmark Berlin10587 Germany Department of Mechanical and Production Engineering iCLIMATE Interdisciplinary Centre for Climate Change Aarhus8000 Denmark Department of Wind and Energy Systems Technical University of Denmark Elektrovej 325 Lyngby2800 Denmark

Road transport is responsible for about a quarter of Europe’s greenhouse gas emissions, making its transformation a crucial part of Europe’s overall decarbonization goals. Current European policies promote decarbonizing the transport sector and passenger car sales show an increased adoption of electric vehicles. Full electrification of land transport will significantly increase the average electricity demand but the use of smart charging and vehicle-to-grid could provide additional flexibility to balance wind and solar generation. In this study, we find cost-optimal transition pathways of the European land transport sector embedded in the sector-coupled open energy model PyPSA-Eur. We consider fossil-fueled, hydrogen-fueled, and electric cars using a 3-hour time resolution for a full year and covering 33 interconnected European countries. We analyze a transition path from 2025 to 2050 under different carbon budgets corresponding to a 1.7 ◦C and 2 ◦C temperature increase. Our results show that rapid electrification of road transport reduces the total system cost, even in the absence of climate targets. We see a clear preference for rapidly decommissioning internal combustion engine vehicles and using electric vehicles in all countries and under all carbon budgets. Allowing smart charging of electric vehicles decreases the total system cost by 1.6% because it reduces the need to install stationary batteries by almost 40%. © 2025, CC BY-NC-SA.

关键词： Vehicle-to-grid

来源：评论

学校读者我要写书评

暂无评论

The Effect of Data Augmentation on the Accuracy of Fish Species Classification Using Deep Learning

The Effect of Data Augmentation on the Accuracy of Fish Spec...

引用

International Conference on Informatics and Computing (ICIC)

作者： Ellya Helmud Catur Edi Widodo Oky Dwi Nurhayati Doctor Information System Postgraduate Diponegoro University Semarang Indonesia Faculty of Information Technology Information Systems Study Program Atma Luhur Institute of Science and Business Pangkalpinang Indonesia Physics Department Faculty of Science and Mathematics Diponegoro University Semarang Indonesia Department of Computer Engineering Faculty of Engineering Diponegoro University Semarang Indonesia

ISBN: (数字)9798331517601

ISBN: (纸本)9798331517618

Fish image classification presents an intriguing challenge in the field of computer vision. This research aims to develop an accurate classification model to differentiate between four different fish species using a convolutional neural network. The dataset used consists of $\mathbf{3 0 1 0}$ fish images, divided into training, validation, and testing sets. The convolutional neural network model was trained both with and without data augmentation. Evaluation results show that the model trained with data augmentation achieved an accuracy of $95 \%$ with a loss value of 0.0983, slightly better than the model without augmentation which achieved an accuracy of $94.56 \%$ with a loss value of $\mathbf{0. 1 7 9 4}$. This indicates that data augmentation techniques are effective in improving model performance, likely because augmentation helps the model generalize better to variations in fish image data. The results of this research demonstrate the significant potential of convolutional neural network for fish image classification tasks. The developed model can serve as a foundation for the development of computer vision-based applications such as automatic fish species identification in fisheries or educational applications. Further research can be conducted by exploring different convolutional neural network architectures, more advanced data augmentation techniques, and larger datasets to improve model performance.

关键词： Deep learning Training Accuracy Computational modeling Computer architecture Fish Data augmentation Data models Convolutional neural networks Image classification

来源：评论

学校读者我要写书评

暂无评论

Preliminary Plan for Data Collection on Stakeholder Information Interdependence in Area-Business Continuity Management: A Case Study of Flood Management in Thai Industrial Zones

Preliminary Plan for Data Collection on Stakeholder Informat...

引用

International Symposium on Communications and Information Technologies (ISCIT)

作者： Penpitcha Arayachookiat Kumpol Saengtabtim Akira Kodaka Natt Leelawat Jing Tang Kenji Watanabe Risk and Disaster Management Program Graduate School Chulalongkorn University Bangkok Thailand Department of Industrial Engineering Faculty of Engineering Chulalongkorn University Bangkok Thailand Graduate School of System Design and Management Keio University Kanagawa Japan Disaster and Risk Management Information Systems Research Unit Chulalongkorn University Bangkok Thailand Faculty of Engineering International School of Engineering Chulalongkorn University Bangkok Thailand Graduate School of Social Engineering Nagoya Institute of Technology Nagoya Japan

ISBN: (数字)9798350353952

ISBN: (纸本)9798350353969

The 2011 flood in Thailand exposed significant vulnerabilities in industrial areas, highlighting the necessity for enhanced disaster risk management through Area-Business Continuity Management (Area-BCM). This preliminary study focuses on identifying how stakeholders rely on information from each other to improve disaster preparedness and response. The research involves systematically identifying Area-BCM stakeholders and designing interview questions, which are evaluated by experts using the Index of Item-Objective Congruence (IOC) to ensure relevance and clarity. All interview questions surpassed the IOC threshold of 0.5, confirming their effectiveness in capturing information interdependencies. Expert feedback led to refinements in the questions, underscoring the importance of tailored data collection. This initial plan provides a critical foundation for understanding information interdependence and highlights the importance of well-designed data collection methods. The findings have significant implications for developing more resilient disaster management strategies in industrial areas, emphasizing the need for precise and relevant stakeholder communication to enhance Area-BCM effectiveness.

关键词： Disasters Disaster management Data collection Information and communication technology Stakeholders Floods Risk management Iterative methods Interviews Resilience

来源：评论

学校读者我要写书评

暂无评论

ProPILE: probing privacy leakage in large language models 23

ProPILE: probing privacy leakage in large language models

引用

Proceedings of the 37th International Conference on Neural Information Processing systems

作者： Siwon Kim Sangdoo Yun Hwaran Lee Martin Gubri Sungroh Yoon Seong Joon Oh Department of Electrical and Computer Engineering Seoul National University NAVER AI Lab University of Luxembourg and Parameter Lab Department of Electrical and Computer Engineering Seoul National University and Interdisciplinary Program in Artificial Intelligence Seoul National University Parameter Lab and Tübingen AI Center University of Tübingen

The rapid advancement and widespread use of large language models (LLMs) have raised significant concerns regarding the potential leakage of personally identifiable information (PII). These models are often trained on vast quantities of web-collected data, which may inadvertently include sensitive personal data. This paper presents ProPILE, a novel probing tool designed to empower data subjects, or the owners of the PII, with awareness of potential PII leakage in LLM-based services. ProPILE lets data subjects formulate prompts based on their own PII to evaluate the level of privacy intrusion in LLMs. We demonstrate its application on the OPT-1.3B model trained on the publicly available Pile dataset. We show how hypothetical data subjects may assess the likelihood of their PII being included in the Pile dataset being revealed. ProPILE can also be leveraged by LLM service providers to effectively evaluate their own levels of PII leakage with more powerful prompts specifically tuned for their in-house models. This tool represents a pioneering step towards empowering the data subjects for their awareness and control over their own data on the web. The demo can be found here: https://***/research/propile

关键词：

来源：评论

学校读者我要写书评

暂无评论

NPRportrait 1.0:A three-level benchmark for non-photorealistic rendering of portraits

引用

Computational Visual Media 2022年第3期8卷 445-465页

作者： Paul L.Rosin Yu-Kun Lai David Mould Ran Yi Itamar Berger Lars Doyle Seungyong Lee Chuan Li Yong-Jin Liu Amir Semmo Ariel Shamir Minjung Son Holger Winnemöller School of Computer Science and Informatics Cardiff UniversityCardiffUK School of Computer Science Carleton UniversityOttawaCanada Department of Computer Science and Technology Tsinghua UniversityBeijingChina Reichman University(the Interdisciplinary Center) HerzliyaIsrael Department of Computer Science and Engineering Pohang University of Science and TechnologyPohangRepublic of Korea Lambda Labs Inc.San FranciscoUSA Hasso Plattner Institute University of PotsdamPotsdamGermany Multimedia Processing Laboratory Samsung Advanced Institute of TechnologySuwonRepublic of Korea Adobe Systems Inc.San JoseUSA

Recently,there has been an upsurge of activity in image-based non-photorealistic rendering(NPR),and in particular portrait image stylisation,due to the advent of neural style transfer(NST).However,the state of performance evaluation in this field is poor,especially compared to the norms in the computer vision and machine learning ***,the task of evaluating image stylisation is thus far not well defined,since it involves subjective,perceptual,and aesthetic *** make progress towards a solution,this paper proposes a new structured,threelevel,benchmark dataset for the evaluation of stylised portrait *** criteria were used for its construction,and its consistency was validated by user ***,a new methodology has been developed for evaluating portrait stylisation algorithms,which makes use of the different benchmark levels as well as annotations provided by user studies regarding the characteristics of the *** perform evaluation for a wide variety of image stylisation methods(both portrait-specific and general purpose,and also both traditional NPR approaches and NST)using the new benchmark dataset.

关键词： non-photorealistic rendering(NPR) image stylization style transfer portrait evaluation benchmark

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：