检索结果-南通市图书馆

Frontiers of Computer Science 2024年第3期18卷 31-42页

作者： Lei LI chengyu WANG Minghui QIU Cen chen Ming GAO Aoying ZHOU Shanghai Engineering Research Center of Big Data Management School of Data Science and EngineeringEast China Normal UniversityShanghai 200062China Alibaba Group Hangzhou 311121China KLATASDS-MOE School of StatisticsEast China Normal UniversityShanghai 200062China

bert is a representative pre-trained language model that has drawn extensive attention for significant improvements in downstream Natural Language Processing(NLP)*** complex architecture and massive parameters bring bert competitive performance but also result in slow speed at model inference *** speed up bert inference,Fastbert realizes adaptive inference with an acceptable drop in accuracy based on knowledge distillation and the early-exit ***,many factors may limit the performance of Fastbert,such as the teacher classifier that is not knowledgeable enough,the batch size shrinkage and the redundant computation of student *** overcome these limitations,we propose a new bert inference method with GPU-Efficient Exit Prediction(GEEP).GEEP leverages the shared exit loss to simplify the training process of Fastbert from two steps into only one step and makes the teacher classifier more knowledgeable by feeding diverse Transformer outputs to the teacher *** addition,the exit layer prediction technique is proposed to utilize a GPU hash table to handle the token-level exit layer distribution and to sort test samples by predicted exit *** this way,GEEP can avoid batch size shrinkage and redundant computation of student *** results on twelve public English and Chinese NLP datasets prove the effectiveness of the proposed *** source codes of GEEP will be released to the public upon paper acceptance.

关键词： bert Fastbert inference acceleration model distillation early exit text classification

Ellipticity on spaces with higher singularities

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Science China Mathematics 2017年第11期60卷 2053-2076页

作者： CHANG Der-chen SCHULZE bert-Wolfgang Department of Mathematics and Statistics Georgetown University Department of Mathematics Fu Jen Catholic University Institute of Mathematics University of Potsdam

We study corner-degenerate pseudo-differential operators of any singularity order and develop ellipticity based on the principal symbolic hierarchy, associated with the stratification of the underlying space. We construct parametrices within the calculus and discuss the aspect of additional trace and potential conditions along lower-dimensional strata.

关键词： pseudo-differential operators operator-valued symbols Fourier and Mellin transforms

Deep Learning and Network Analysis:Classifying and Visualizing Geologic Hazard Reports

在线全文

学校读者我要写书评

暂无评论

Journal of Earth Science 2024年第4期35卷 1289-1303页

作者： Wenjia Li Liang Wu Xinde Xu Zhong Xie Qinjun Qiu Hao Liu Zhen Huang Jianguo chen School of Geography and Information Engineering China University of GeosciencesWuhan430078China Key Laboratory of Geological Survey and Evaluation of Ministry of Education China University of GeosciencesWuhan430078China School of Computer Science China University of GeosciencesWuhan430078China Key Laboratory of Urban Land Resources Monitoring and Simulation Ministry of Natural ResourcesShenzhen518034China Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering China Three Gorges UniversityYichang443002China Wuhan Geomatics Institute Wuhan430074China Faculty of Earth Resources China University of GeosciencesWuhan430074China

If progress is to be made toward improving geohazard management and emergency decision-making,then lessons need to be learned from past geohazard information.A geologic hazard report provides a useful and reliable source of information about the occurrence of an event,along with detailed information about the condition or factors of the *** such reports,however,can be a challenging process because these texts are often presented in unstructured long text formats,and contain rich specialized and detailed *** text classification is commonly used to mine disaster text data in open domains(e.g.,news and microblogs).But it has limitations to performing contextual long-distance dependencies and is insensitive to discourse *** deficiencies are most obviously exposed in long text ***,this paper uses the bidirectional encoder representations from Transformers(bert),to model long ***,utilizing a softmax layer to automatically extract text features and classify geohazards without manual *** latent Dirichlet allocation(LDA)model is used to examine the interdependencies that exist between causal variables to visualize *** proposed method is useful in enabling the machine-assisted interpretation of text-based ***,it can help users visualize causes,processes,and other geohazards and assist decision-makers in emergency responses.

关键词： geologic hazard network analysis latent dirichlet allocation text classification deep learning

中国城市创业情感的时空差异及其对企业家精神的影响

在线全文

学校读者我要写书评

暂无评论

地理研究 2024年第1期43卷 31-50页

作者：陈广平王琛张佳浙江大学地球科学学院杭州310058 多伦多大学地理与规划学院加拿大多伦多M5S1A1 墨尔本大学建筑与规划学院澳大利亚墨尔本3010

创业情感是个体对创业的主观体验和感受,反映了社会对创业者及其创业活动的认可和接受程度。本文基于2014—2018年新浪微博“创业”话题的36.82万条文本数据,运用自然语言分析技术、文本数据分析与相关空间分析方法,在定量测度中国192... 详细信息

创业情感是个体对创业的主观体验和感受,反映了社会对创业者及其创业活动的认可和接受程度。本文基于2014—2018年新浪微博“创业”话题的36.82万条文本数据,运用自然语言分析技术、文本数据分析与相关空间分析方法,在定量测度中国192个城市的创业情感基础上,分析了正、负面两类创业情感的总体特征和时空差异,采用面板固定效应模型探究了创业情感对地区企业家精神的影响。研究发现:①个体创业情感主要受创业社会网络、创业体验、创业动机、创业环境、创业方式等因素的影响,其中“朋友”的词频最高,表明非亲属的熟人关系是影响个体创业情绪的关键要素。语义网络显示具有正面情感的创业者更关注投资机会和注重发挥创业平台的作用,而具有负面情绪的个体可能对房价、租金等与创业成本相关的因素更为敏感。②在城市尺度,正面的创业情感呈现“南高北低”“沿海高内陆低”的空间格局。随着时间推移,中国城市居民创业情感总体上趋于弱化且出现了空间“极化”现象。特别地,珠三角地区的深圳、广州、珠海和长三角地区的杭州、南京、苏州、宁波等城市始终是创业者的“乐园”。③模型结果显示,正面的创业情绪能够提振企业家精神,地区正面创业情感比例越高,越有利于促进地区创业水平的提升。政府的创业干预措施可以正向调节创业情感并有助于提振当地企业家精神,而高房价给创业者带来的负面创业情绪会显著削弱企业家精神。本研究基于大数据分析,定量论证了创业情感和社会态度对企业家精神的影响,可为区域创业政策调整、地区良好创业氛围的营造以及个体创业情绪的调节提供参考。

关键词：创业情感企业家精神新浪微博 bert模型非正式制度情感分析

DPAL-bert:A Faster and Lighter Question Answering Model

在线全文

学校读者我要写书评

暂无评论

Computer Modeling in Engineering & Sciences 2024年第10期141卷 771-786页

作者： Lirong Yin Lei Wang Zhuohang Cai Siyu Lu Ruiyang Wang Ahmed AlSanad Salman A.AlQahtani Xiaobing chen Zhengtong Yin Xiaolu Li Wenfeng Zheng Department of Geography and Anthropology Louisiana State UniversityBaton RougeLA 70803USA School of Automation University of Electronic Science and Technology of ChinaChengdu610054China College of Computer and Information Sciences King Saud UniversityRiyadh11574Saudi Arabia School of Electrical and Computer Engineering Louisiana State UniversityBaton RougeLA 70803USA College of Resources and Environmental Engineering Guizhou UniversityGuiyang550025China School of Geographical Sciences Southwest UniversityChongqing400715China

Recent advancements in natural language processing have given rise to numerous pre-training language models in question-answering ***,with the constant evolution of algorithms,data,and computing power,the increasing size and complexity of these models have led to increased training costs and reduced *** study aims to minimize the inference time of such models while maintaining computational *** also proposes a novel Distillation model for PAL-bert(DPAL-bert),specifically,employs knowledge distillation,using the PAL-bert model as the teacher model to train two student models:DPAL-bert-Bi and *** research enhances the dataset through techniques such as masking,replacement,and n-gram sampling to optimize knowledge *** experimental results showed that the distilled models greatly outperform models trained from *** addition,although the distilled models exhibit a slight decrease in performance compared to PAL-bert,they significantly reduce inference time to just 0.25%of the *** demonstrates the effectiveness of the proposed approach in balancing model performance and efficiency.

关键词： DPAL-bert question answering systems knowledge distillation model compression bert Bi-directional long short-term memory(BiLSTM) knowledge information transfer PAL-bert training efficiency natural language processing

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

基于规则与学习的变异技术对比研究

软件学报 2024年第7期35卷 3093-3114页

作者：贡志豪陈逸洲陈俊洁郝丹高可信软件技术教育部重点实验室(北京大学) 北京100871 北京大学计算机学院北京100871 天津大学智能与计算学部天津300350

变异测试是一种有效的软件测试技术,通过生成变异体来模拟软件缺陷,帮助提升现有测试套件的缺陷检测能力.变异体的质量对于变异测试的有效性具有显著影响.传统的变异测试方法通常采用人工设计的基于语法规则的变异算子生成变异体,并已... 详细信息

变异测试是一种有效的软件测试技术,通过生成变异体来模拟软件缺陷,帮助提升现有测试套件的缺陷检测能力.变异体的质量对于变异测试的有效性具有显著影响.传统的变异测试方法通常采用人工设计的基于语法规则的变异算子生成变异体,并已取得一定的研究成果.近年来,许多研究开始结合深度学习技术,通过学习开源项目历史代码生成变异体.目前,该新方法在变异体生成方面取得了初步的成果.基于语法规则和基于学习的两种变异技术,其机理不同,但其目标均是通过生成变异体来提高测试套件的缺陷检测能力,因此,全面比较这两种变异技术对于变异测试及其下游任务至关重要.针对这一问题,设计实现了一项针对基于语法规则和基于学习的变异技术的实证研究,旨在了解不同机理的变异技术在变异测试任务上的性能以及生成的变异体在程序语义上的差异性.具体地,以Defect4J v1.2.0数据集为实验对象,比较以MAJOR和PIT为代表的基于语法规则的变异技术和以DeepMutation、μbert和LEAM为代表的基于深度学习的变异技术.实验结果表明:基于规则与学习的变异技术均可有效支持变异测试实践,但MAJOR的测试效果最优,能够检测85.4%的真实缺陷.在语义表示上,MAJOR具有最强的语义代表能力,基于其构造的测试套件能够杀死其余变异技术生成的超过95%占比的变异体.在缺陷表征上,两类技术均具有独特性.

关键词：变异测试变异分析实证研究缺陷检测

Upcycling biomass waste into Fe single atom catalysts for pollutant control

在线全文

学校读者我要写书评

暂无评论

Journal of Energy Chemistry 2022年第6期31卷 282-291,I0008页

作者： Xin Li Kang Hu Yizhe Huang Qingqing Gu Yuwen chen Bing Yang Rongliang Qiu Wenhao Luo bert M.Weckhuysen Kai Yan Guangdong Provincial Key Laboratory of Environmental Pollution Control and Remediation Technology School of Environmental Science and EngineeringSun Yat-Sen UniversityGuangzhou 510275GuangdongChina CAS Key Laboratory of Science and Technology on Applied Catalysis Dalian Institute of Chemical PhysicsChinese Academy of SciencesDalian 116023LiaoningChina College of Natural Resources and Environment South China Agricultural UniversityGuangzhou 510275GuangdongChina Inorganic Chemistry and Catalysis Group Debye Institute for Nanomaterials ScienceUtrecht UniversityUtrecht 3584 CGNetherlands

Contaminants of heavy metals and antibiotics, which are frequently detected in water, soil and food chains with increasing prevalence in our current society, can cause potential harm to human health and disrupt human ecosystem irreversibly. Herein, we have successfully utilized biomass waste ferns contaminated by iron mines, to fabricate a first-of-its-kind high-performance class of Fe single-atom catalysts(FeSAC) by a facile pyrolysis. The optimal FeSAC-800 shows an excellent efficiency in the fastphotocatalytic degradation of six types of quinolone antibiotics(e.g., norfloxacin, levofloxacin, ciprofloxacin, enrofloxacin, lomefloxacin, flumequine) in 1 h under the simulated natural light irradiation. Based on advanced characterization, a well-defined structure of FeN_(4), confined in the porous carbon is elaborated for the FeSAC-800. Mechanism of the photodegradation is via a Fenton-like oxidation process whereas the reactive oxygen species play a key role. These findings open a new avenue for efficient, sustainable utilization of biomass waste in pollutant control.

关键词： Single-atom catalysts Photocatalysis Biomass waste Quinolone antibiotics Sustainability

基于文本数据增强的中文水稻育种问句命名实体识别

在线全文

学校读者我要写书评

暂无评论

农业机械学报 2024年第8期55卷 333-343页

作者：牛培宇侯琛中国农业大学信息与电气工程学院北京100083 北京大学大数据分析与应用技术国家工程实验室北京100871 北京大学长沙计算与数字经济研究院长沙410205

针对现有水稻育种问答系统存在数据管理水平低、知识粒度大,水稻育种领域缺乏用于命名实体识别的标注数据、人工标注成本高等问题,提出了一种基于文本数据增强的方法来识别水稻育种问句的命名实体,通过构建水稻育种知识图谱,对水稻育种... 详细信息

针对现有水稻育种问答系统存在数据管理水平低、知识粒度大,水稻育种领域缺乏用于命名实体识别的标注数据、人工标注成本高等问题,提出了一种基于文本数据增强的方法来识别水稻育种问句的命名实体,通过构建水稻育种知识图谱,对水稻育种问句中的大类命名实体进行分类,从而增强实体边界,降低知识粒度。针对水稻育种数据标注成本高导致命名实体识别性能不佳的难点,通过在bert-BILSTM-CRF模型中引入数据增强层,提出了DA-bert-BILSTM-CRF模型。实验以标注的水稻育种问句为训练数据,将所提出的模型与其他基线模型进行比较。结果表明,本文方法在水稻育种问句中命名实体识别的单类别识别任务和整体识别任务上均优于其他方法,其中单类别识别精确率达到94.26%,F1值达到93.32%;整体识别精确率达到93.86%,F1值达到93.34%。

关键词：水稻育种问答系统命名实体识别文本数据增强知识图谱

基于Roberta和集中注意力机制的营商政策多标签分类

同方期刊数据库博看期刊评论

在线全文

学校读者我要写书评

暂无评论

计算机应用 2024年第S01期44卷 44-48页

作者：陈昊飏南京大学计算机科学与技术系南京210023

为了满足营商政策多标签分类的社会需求,解决使用擅长文本分类、但输入受限的大语言预训练模型进行长文本分类的难题,提出一种基于Roberta模型和集中注意力机制的方法,更好地提取语义集中区域的信息表征,对营商政策文本进行有效的多标... 详细信息

为了满足营商政策多标签分类的社会需求,解决使用擅长文本分类、但输入受限的大语言预训练模型进行长文本分类的难题,提出一种基于Roberta模型和集中注意力机制的方法,更好地提取语义集中区域的信息表征,对营商政策文本进行有效的多标签分类。首先,对数据清洗和分析后,得到一定的先验知识:营商政策文本的语义表征集中在文本标题与开篇部分。其次,在文本输入层和向量表示层中,构建集中注意力机制对文本和向量进行处理,增强模型在训练中对语义集中区域的注意力,提高模型信息表征提取能力,优化长文本分类的效果。实验中爬取政府公开的营商政策文本作为数据集,实验结果表明,营商政策长文本分类的准确率可达0.95,Micro-F1值可达0.91,同时对比实验结果显示,融合Roberta和集中注意力机制进行营商政策长文本多标签分类比其他模型效果更好。

关键词：多标签分类长文本营商政策 Roberta 预训练模型注意力机制

同方期刊数据库