检索结果-内蒙古大学图书馆

6～(th) international conference on Spoken language processing

作者： Yumi WAKITA Kenji MATSUI Foshinori SAGISAKA Advanced Technology Research Laboratories Matsushita Electric Industrial Co. Ltd. Kyoto Japan Graduate School of Science and Technology Kobe University Kobe Japan Advanced Technology Research Laboratories Matsushita Electric Industrial Co. Ltd. Kyoto Japan ATR Spoken Language Translation Research Laboratories Kyoto Japan Graduate School of Science and Technology Kobe University Kobe Japan

For robust speech translation, we propose a new language translation method in which speech recognition results are mapped to example sentences using keywords. In this method, the keyword clustering is used to cope with recognition crrors and the wide variety of words that do not appear in the training corpus. Initial classes defined using only thesaurus are redefined by using the dependency between the keywords in limited number of example sentences. the effectiveness of our keyword clustering method is confirmed through example sentence search experiments. these experiments were done using keyword sets of (a) different sentences including keywords not in the example sentences and (b) recognition results those sentences in which recognition errors were obtained. Compared with the search method which uses keyword sets defined by using only a thesaurus, our proposed method offered improved search error rates.

关键词： domain LIS Fine keyword clustering using a thesaurus and example sentences for speech translation

来源：评论

学校读者我要写书评

暂无评论

Vision AI: Enhancing Ophthalmic Care with advanced Detection Technologies

Vision AI: Enhancing Ophthalmic Care with Advanced Detection...

引用

Advancements in Computing (ICAC), international conference on

作者： S. M. K. S. Siriwardana S. N. W. Gunasekara W. U. Piyumantha S. P. thilakaratne D. R. Wijendra D. I. de Silva Faculty of Computing Sri Lanka Institute of Information Technology Malabe Sri Lanka

ISBN: (数字)9798331517878

ISBN: (纸本)9798331517885

this study presents Vision AI, an advanced diagnostic framework that utilizes hybrid deep learning models for accurate detection of eye diseases, specifically Glaucoma, Macular Hole, Central Serous Retinopathy, and Drusen. Traditional ophthalmic diagnostics often suffer from human error and inefficiencies, emphasizing the need for automated solutions. Vision AI addresses these challenges by integrating multiple model architectures, such as VGG16, ResNet, V-Net, and YOLOv8x, optimizing diagnostic accuracy and generalizability. A carefully curated dataset of retinal images was used, supported by extensive data augmentation and hyperparameter tuning to enhance model robustness. this approach significantly improved detection capabilities, achieving over 95% specificity and sensitivity across diseases. Additionally, Vision AI includes a Large language Model for providing tailored health recommendations, making the system highly practical for clinical use. the framework is deployed in a scalable microservice-based web application, ensuring modularity, security, and real-time diagnostics, thus improving accessibility for healthcare providers. Vision AI showcases the potential of hybrid deep learning systems to transform ophthalmic care, laying a foundation for AI-assisted, patient-centered healthcare advancements.

关键词： Deep learning Glaucoma Technological innovation Accuracy Sensitivity Transforms Retina Robustness Security Tuning

来源：评论

学校读者我要写书评

暂无评论

Design and Implementation of A Cache Adapted to the LoongArch Architecturert 22

Design and Implementation of A Cache Adapted to the LoongArc...

引用

Proceedings of the 2022 6th international conference on Electronic information technology and Computer Engineering

作者： Lin Chen Xiao Ma Xiang Ji Ziyan Wang School of Information Science Beijing Language and Culture University China

ISBN: (纸本)9781450397148

A cache adapted to the LoongArch architecture is designed and implemented using the FPAG platform. A simple and performance balanced Cache module is designed and implemented. the design of the read and write data paths of the Cache module is given, and a state machine is designed and implemented for the logical control of the Cache module. the interaction interface between the Cache and the CPU is designed. the CPU pipeline adaptation method for Cache in hit and unhit cases is given. the processing flow of UnCache access is given. Cache-related instructions in the LoongArch instruction set are implemented. the test results show that the scheme meets the design requirements. Cache is simple and effective, and is suitable as a case study for initial understanding of Cache functionality and structure.

关键词： LoongArch Cache FPGA CPU

来源：评论

学校读者我要写书评

暂无评论

A semiring-based trace semantics for processes with applications to information leakage analysis

IFIP Advances in Information and Communication Technology

引用

IFIP Advances in information and Communication technology 2010年 323 AICT卷 340-354页

作者： Boreale, Michele Clark, David Gorla, Daniele Dipartimento di Sistemi e Informatica Università di Firenze Italy Department of Computer Science King's College London United Kingdom Dipartimento di Informatica Sapienza Universitàdi Roma Italy

ISBN: (纸本)3642152392

We propose a framework for reasoning about program security building on language-theoretic and coalgebraic concepts. the behaviour of a system is viewed as a mapping from traces of high (unobservable) events to low (observable) events: the less the degree of dependency of low events on high traces, the more secure the system. We take the abstract view that low events are drawn from a generic semiring, where they can be combined using product and sum operations;throughout the paper, we provide instances of this framework, obtained by concrete instantiations of the underlying semiring. We specify systems via a simple process calculus, whose semantics is given as the unique homomorphism from the calculus into the set of behaviours, i.e. formal power series, seen as a final coalgebra. We provide a compositional semantics for the calculus in terms of rational operators on formal power series and show that the final and the compositional semantics coincide. © IFIP international Federation for information processing 2010.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Constructing a Medical Domain Functional Knowledge Graph with Large language Models

Constructing a Medical Domain Functional Knowledge Graph wit...

引用

IEEE international conference on Power, Intelligent Computing and Systems (ICPICS)

作者： Lintian Ji Shuxin Du Yi Qiu Hongyi Xu Xiaorui Guo Huzhou Key Laboratory of Intelligent Sensing and Optimal Control for Industrial Systems School of Engineering Huzhou University Huzhou China Huzhou Key Laboratory of Intelligent Sensing and Optimal Control for Industrial Systems School of Engineering Huzhou University Huzhou China

ISBN: (数字)9798350374315

ISBN: (纸本)9798350374322

the integration of medicine and artificial intelligence is creating novel opportunities to gather, analyze, and generate groundbreaking medical insights from the vast expanse of medical literature. Although these advancements hold great promise, difficulties including manual annotation and precise data retrieval, and transparency persist. advanced natural language processing (NLP) techniques and large language models (LLMs) have shown potential in addressing these issues. this paper introduces the Medical Knowledge Graph (MedKG), a comprehensive and multidisciplinary medical knowledge graph. By utilizing advanced NLP algorithms, MedKG extracts millions of entities and converts them into triples from a corpus of superior medical research papers published over the last decade. It categorizes unstructured medical information into nine distinct tags, including disease names, treatments, diagnoses, symptoms, genetics, cellular structures of diseases, patient data, medical procedures, and treatment applications, while seamlessly integrating the digital object identifier (DOI) of each paper. As the latest structured database of medical knowledge, MedKG serves as a powerful tool for accelerating medical research and lays the groundwork for constructing a more extensive medical knowledge graph using full-text medical articles. Moreover, our research establishes a foundation for practical knowledge management systems based on text mining, applicable not only to complex medical systems but also to other professional domains.

关键词： Text mining Annotations Large language models Pipelines Knowledge graphs Prediction algorithms Natural language processing Object recognition Medical diagnostic imaging Diseases

来源：评论

学校读者我要写书评

暂无评论

3D Hand Pose Estimation Based on External Attention

3D Hand Pose Estimation Based on External Attention

引用

Natural language processing (ICNLP), international conference on

作者： Shoukun Li Xiaoying Pan Beibei Wang Jialong Gao School of Computer Science and Technology Xi'an University of Posts and Telecommunications Xi’an China Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing Xi'an University of Posts and Telecommunications Xi'an China

ISBN: (数字)9798350349115

ISBN: (纸本)9798350349122

3D hand pose for a single depth image is an important topic in computer vision and human-computer interaction, and although significant progress has been made in this field in recent years, there is still room for improvement in accuracy for some specific application scenarios. To address this problem, a 3D hand pose estimation algorithm based on external attention is proposed. First, the target features are extracted by an hourglass network; then, a HEM (Hard Example Mining) loss based on a mean-variance loss function is proposed, which firstly calculates the L2 loss values of all N keypoints, and then sorts these loss values, and back-propagates the gradient only to the first m loss values. Meanwhile, external attention is introduced to enhance the ability to perceive the global information of the target, and the recognition ability of the features is improved by giving the features different influences through the attention weights. Experimental results show that the algorithm achieves an average distance error of 5.42 mm on the ICVL dataset and 7.11 mm on the MSRA dataset, which further improves the detection performance of the 3D hand pose estimation algorithm.

关键词： Human computer interaction Computer vision three-dimensional displays Accuracy Target recognition Pose estimation Feature extraction Prediction algorithms Natural language processing

来源：评论

学校读者我要写书评

暂无评论

Enhanced Document Segmentation at Paragraph Level

Enhanced Document Segmentation at Paragraph Level

引用

Electronics and Communication, Network and Computer technology (ECNCT), international conference on

作者： Jiafu Ma Sai Han Shiwei Ye Guangquan Wang University of Chinese Academy of Sciences Beijing China China Unicom Research Institute Beijing China

ISBN: (数字)9798350366143

ISBN: (纸本)9798350366150

the rapid growth of digital communications has led to an exponential increase in textual data, necessitating efficient methods for document segmentation. the readability can be improved by segmenting long documents into semantically coherent sections. Downstream Natural language processing (NLP) tasks, such as information retrieval and document summarization, can also be greatly assisted. However, previous studies have predominantly focused on sentence-level segmentation, which often failed to utilize the inherent paragraph structure of documents effectively. In this paper, a novel approach to document segmentation that leverages the characteristics of paragraphs is proposed. Pretrained language models (PLMs) are employed to identify and segment distinct topics within documents by analyzing the structure and semantic coherence between paragraphs. To improve the model performance, two new auxiliary tasks are integrated into the training process, providing additional signals that helps in better understanding context coherence and document structure. these auxiliary tasks are designed to align with the primary objective of segmentation while enhancing the model's capability to generalize across various types of documents. A novel dataset with approximately 10,000 documents focused on communications domain, characterized by detailed paragraph and topic structure, is crawled for training and evaluating. A diverse set of other domain specific documents were also evaluated with our model. It is demonstrated that the results of integrating auxiliary tasks and leveraging paragraph-level features significantly outperform baseline models.

关键词： Training Analytical models Computational modeling Semantics Coherence information retrieval Digital communication Natural language processing Communication networks Context modeling

来源：评论

学校读者我要写书评

暂无评论

Improvement of Timing Synchronization Algorithm for FBMC-OQAM System Based on Training Sequence

Improvement of Timing Synchronization Algorithm for FBMC-OQA...

引用

Natural language processing (ICNLP), international conference on

作者： Xihai Xie Ruichao Ning School of Communications and Information Engineering Xi’an University of Posts & Telecommunications Xi’an China

ISBN: (数字)9798350349115

ISBN: (纸本)9798350349122

In the context of 5G with high rate transmission, the filter bank multicarrier-offset orthogonal amplitude modulation (FBMC-OQAM) technique has attracted much attention as an alternative to OFDM technology because the FBMC system reduces the wastage of spectral resources and has better time-frequency focusing characteristics, as well as better robustness to channel frequency selective fading. However, the inter-symbol overlap and interference caused by this modulation can make timing synchronization difficult. To address this problem, this paper designs a special training sequence based on existing synchronization algorithms, which exploits the conjugate symmetry property of the FBMC system for timing synchronization, and consists of four symbols, the first two of which are computed to have conjugate symmetry in the time domain; and the last two of which are zero-valued. the length of the correlation sequence is enlarged compared to the traditional algorithm, and the number of symbols used for training is increased. theoretical analysis and simulation results show that the proposed algorithm has higher timing accuracy and better synchronization performance.

关键词： Training Time-frequency analysis Analytical models Accuracy OFDM Simulation Symbols Interference Synchronization Time-domain analysis

来源：评论

学校读者我要写书评

暂无评论

Enhancing Code Completion with Position Embedding of Abstract Syntax Tree

Enhancing Code Completion with Position Embedding of Abstrac...

引用

Frontiers technology of information and Computer (ICFTIC), IEEE international conference on

作者： Jiahao Li Junyu Lu Linbo Zhu AHU-IAI AI Joint Laboratory Hefei Comprehensive National Science Center Anhui University Institute of Artificial Intelligence Hefei China School of Data Science University of Science and Technology of China Hefei China Institute of Artificial Intelligence Hefei Comprehensive National Science Center Hefei China

ISBN: (数字)9798331541750

ISBN: (纸本)9798331541767

Code completion is a crucial feature in integrated development environments (IDEs) that enhances both the efficiency and experience of developers. By analyzing the context of preceding code, real-time suggestions can be provided, significantly aiding developers in their work. Recent research has highlighted that the abstract syntax tree (AST) contains rich syntactic and structural information, which can assist deep learning models in better understanding code. As a result, AST-based code completion methods have garnered substantial attention. However, existing approaches often overlook the positional information of nodes within the AST, limiting the model's ability to capture structural relationships in the code. this paper explores the positional relationships between AST nodes and proposes a tree structure position embedding method tailored to the AST. Experimental results validate that this position embedding significantly improves the performance of code completion.

关键词： Deep learning Codes Limiting Computational modeling Syntactics Real-time systems Natural language processing

来源：评论

学校读者我要写书评

暂无评论

An Energy-Efficient Architecture of Approximate Softmax Functions for Transformer in Edge Computing

An Energy-Efficient Architecture of Approximate Softmax Func...

引用

Electronics and Electrical Engineering technology (EEET), international conference on

作者： Suhang Li Bo Yin Hao Zhang Information Science and Engineering Ocean University of China Qingdao China

ISBN: (数字)9798350395594

ISBN: (纸本)9798350395600

Recently, Transformer models have provided better accuracy than traditional models in fields such as computer vision (CV) and natural language processing (NLP). However, compared with traditional convolutional neural networks (CNNs), a large number of softmax function calculations in Transformer models involve expensive exponential and division calculations, resulting in huge consumption of computing and storage resources and waste of power consumption, which makes Transformer networks unable to be effectively applied to edge computing. To solve this problem, in this paper an approximate softmax computation architecture is proposed. Compared with the classic base-e softmax function or the base-2 softmax function proposed in recent years, our proposed design uses simpler shift and fixed-point addition operations to replace complex exponential and division operations, and thus significantly reduces the resource consumption while maintaining a high computation accuracy. the proposed architecture can effectively accelerate the Softmax computation of Transformer models in edge computing. Experimental results show that for 16-bit Softmax computation, the proposed design can achieve less resource occupation and energy consumption. Vision transformer(ViT) models are also used to verify the effectiveness of the proposed method in real transformer models.

关键词： Computer vision Accuracy Power demand Computational modeling Computer architecture Transformers Natural language processing

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：