检索结果-南通市图书馆

同方期刊数据库

面向人工智能的传统音乐标注数据集构建研究——兼论多模态胡琴标注数据集的建设与应用

在线全文

学校读者我要写书评

暂无评论

中央音乐学院学报 2024年第2期 66-83页

作者：张宇孙茂松中央音乐学院音乐人工智能与音乐信息科技系清华大学人工智能研究院中央音乐学院

当前人工智能在音乐方面的研究和应用主要集中在西方古典音乐和流行音乐等领域。以世界各地的传统音乐为对象的相关研究仍然处于初级阶段,其中一个主要原因是缺乏高质量的数据集和标注规范。文章从数据的内容、标注和应用等方面,探讨了... 详细信息

当前人工智能在音乐方面的研究和应用主要集中在西方古典音乐和流行音乐等领域。以世界各地的传统音乐为对象的相关研究仍然处于初级阶段,其中一个主要原因是缺乏高质量的数据集和标注规范。文章从数据的内容、标注和应用等方面,探讨了传统音乐标注数据集的发展现状。基于对当前研究的深入分析,文章提出了面向人工智能的音乐标注数据集的构建原则,并针对传统音乐的特点加以阐述。在此基础上,以第一个中国胡琴音乐标注数据集(CCOM-HuQin)为例介绍了数据集在人工智能领域的研究成果和应用前景。文章强调了人工智能应用于传统音乐的意义不仅在于辅助创作和表演,更是为了促进传统音乐的保护、传承和发展。

关键词：人工智能标注数据集音乐信息检索传统音乐胡琴

零售商公平关切下闭环供应链产品升级生产与协调机制研究

在线全文

学校读者我要写书评

暂无评论

物流技术 2024年第2期43卷 128-138页

作者：邢光军孙茂松南京邮电大学管理学院江苏南京210003

研究了由一个制造商和零售商组成的产品升级闭环供应链的生产协调问题。引入公平关切系数后,分别讨论了产品升级闭环供应链集中与分散模型下的最优生产决策,分析零售商公平关切行为、产品升级前后性能差距对产品升级闭环供应链利润的影... 详细信息

研究了由一个制造商和零售商组成的产品升级闭环供应链的生产协调问题。引入公平关切系数后,分别讨论了产品升级闭环供应链集中与分散模型下的最优生产决策,分析零售商公平关切行为、产品升级前后性能差距对产品升级闭环供应链利润的影响,并提出“定价契约”对产品升级闭环供应链利润进行协调。研究表明,制造商需要合理控制升级前后产品性能差距才能使得升级前后产品拥有更好的市场;随着零售商公平关切行为程度的提升,制造商的利润会向零售商发生转移;控制升级性能差距不变,产品不断地推陈出新会提高供应链的整体收益;采用“定价契约”可以协调制造商和零售商之间的收益,实现帕累托优化。

关键词：闭环供应链产品升级公平关切供应链协调

Enriching the Transfer Learning with Pre-Trained Lexicon Embedding for Low-Resource Neural Machine Translation

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

Tsinghua Science and Technology 2022年第1期27卷 150-163页

作者： Mieradilijiang Maimaiti Yang Liu Huanbo Luan maosong sun the Institute for Artificial Intelligence Beijing National Research Center for Information Science and TechnologyDepartment of Computer Science and TechnologyTsinghua UniversityBeijing 100084China the Beijing Academy of Artificial Intelligence Beijing Advanced Innovation Center for Language ResourcesBeijing 100084China

Most State-Of-The-Art(SOTA) Neural Machine Translation(NMT) systems today achieve outstanding results based only on large parallel *** large-scale parallel corpora for high-resource languages is easily ***,the translation quality of NMT for morphologically rich languages is still unsatisfactory,mainly because of the data sparsity problem encountered in Low-Resource Languages(LRLs).In the low-resource NMT paradigm,Transfer Learning(TL) has been developed into one of the most efficient *** is difficult to train the model on high-resource languages to include the information in both parent and child models,as well as the initially trained model that only contains the lexicon features and word embeddings of the parent model instead of the child languages *** this work,we aim to address this issue by proposing the language-independent Hybrid Transfer Learning(HTL) method for LRLs by sharing lexicon embedding between parent and child languages without leveraging back translation or manually injecting ***,we train the High-Resource Languages(HRLs) as the parent model with its ***,we combine the parent and child language pairs using the oversampling method to train the hybrid model initialized by the previously parent ***,we fine-tune the morphologically rich child model using a hybrid ***,we explore some exciting discoveries on the original TL *** results show that our model consistently outperforms five SOTA methods in two languages Azerbaijani(Az) and Uzbek(Uz).Meanwhile,our approach is practical and significantly better,achieving improvements of up to 4:94 and 4:84 BLEU points for low-resource child languages Az ! Zh and Uz ! Zh,respectively.

关键词： artificial intelligence natural language processing neural network machine translation low-resource languages transfer learning

Red Alarm for Pre-trained Models:Universal Vulnerability to Neuron-level Backdoor Attacks

在线全文

学校读者我要写书评

暂无评论

Machine Intelligence Research 2023年第2期20卷 180-193页

作者： Zhengyan Zhang Guangxuan Xiao Yongwei Li Tian Lv Fanchao Qi Zhiyuan Liu Yasheng Wang Xin Jiang maosong sun Department of Computer Science and Technology Tsinghua UniversityBeijing 100084China Institute for Artificial Intelligence Tsinghua UniversityBeijing 100084China Beijing National Research Center for Information Science and Technology Beijing 100084China Huawei Noah′s Ark Laboratory Hong Kong 999077China

The pre-training-then-fine-tuning paradigm has been widely used in deep *** to the huge computation cost for pre-training,practitioners usually download pre-trained models from the Internet and fine-tune them on downstream datasets,while the downloaded models may suffer backdoor *** from previous attacks aiming at a target task,we show that a backdoored pre-trained model can behave maliciously in various downstream tasks without foreknowing task *** can restrict the output representations(the values of output neurons)of trigger-embedded samples to arbitrary predefined values through additional training,namely neuron-level backdoor attack(NeuBA).Since fine-tuning has little effect on model parameters,the fine-tuned model will retain the backdoor functionality and predict a specific label for the samples embedded with the same *** provoke multiple labels in a specific task,attackers can introduce several triggers with predefined contrastive *** the experiments of both natural language processing(NLP)and computer vision(CV),we show that NeuBA can well control the predictions for trigger-embedded instances with different trigger *** findings sound a red alarm for the wide use of pre-trained ***,we apply several defense methods to NeuBA and find that model pruning is a promising technique to resist NeuBA by omitting backdoored neurons.

关键词： Pre-trained language models backdoor attacks transformers natural language processing(NLP) computer vision(CV)

Unified Framework of Performing Chinese Word Segmentation and Part-of-Speech Tagging

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

China Communications 2012年第3期9卷 1-9页

作者： Zhang Kaixu sun maosong Tsinghua Univ State Key Lab Intelligent Technol & Syst Dept Comp Sci & Technol Beijing 100084 Peoples R China

The paper proposes a unified framework to combine the advantages of the fast one-at-a-time approach and the high-performance all-at-once approach to perform Chinese Word Segmentation(CWS) and Part-of-Speech(PoS) *** this framework,the input of the PoS tagger is a candidate set of several CWS results provided by the CWS *** widely used one-at-a-time approach and all-at-once approach are two extreme cases of the proposed candidate-based *** on Penn Chinese Treebank 5 and Tsinghua Chinese Treebank show that the generalized candidate-based approach outperforms one-at-a-time approach and even the all-at-once *** candidate-based approach is also faster than the time-consuming all-at-once *** authors compare three different methods based on sentence,words and character-intervals to generate the candidate *** turns out that the word-based method has the best performance.

关键词：框架词性标注汉语分词表演中文分词水煤浆候选人日期设置

Mining the interests of Chinese microbloggers via keyword extraction

在线全文

学校读者我要写书评

暂无评论

Frontiers of Computer Science 2012年第1期6卷 76-87页

作者： Zhiyuan LIU Xinxiong CHEN maosong sun Department of Computer Science and Technology State Key Lab on Intelligent Technology and SystemsNational Lab for Information Science and TechnologyTsinghua UniversityBeijing 100084China

Microblogging provides a new platform for com- municating and sharing information among Web users. Users can express opinions and record daily life using microblogs. Microblogs that are posted by users indicate their interests to some extent. We aim to mine user interests via keyword extraction from microblogs. Traditional keyword extraction methods are usually designed for formal documents such as news articles or scientific papers. Messages posted by mi- croblogging users, however, are usually noisy and full of new words, which is a challenge for keyword extraction. In this paper, we combine a translation-based method with a frequency-based method for keyword extraction. In our ex- periments, we extract keywords for microblog users from the largest microblogging website in China, Sina Weibo. The re- suits show that our method can identify users＇ interests accu- rately and efficiently.

关键词： microblogging Sina Weibo Chinese keywordextraction user interests.

A Computational Model of Concept Generalization in Cross-Modal Reference

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

Tsinghua Science and Technology 2011年第2期16卷 113-120页

作者： Patrick McCrae Wolfgang Menzel maosong sun CINACS Graduate Research Group Department of Informatics Hamburg University 22527 Hamburg Germany Natural Language Systems Group Department of Informatics Hamburg University 22527 Hamburg Germany Department of Computer Science Tsinghua University Beijing 100084 China

Cross-modal interactions between visual understanding and linguistic processing substantially contribute to the remarkable robustness of human language *** argue that the formation of cross-modal referential links is a prerequisite for the occurrence of cross-modal interactions between vision and *** this paper we examine a computational model for a cross-modal reference formation with respect to its robustness against conceptual underspecification in the visual *** investigation is motivated by the fact that natural systems are well capable of establishing a cross-modal reference between modalities with different degrees of conceptual *** the investigated model,conceptually underspecified context information continues to drive the syntactic disambiguation of verb-centered syntactic ambiguities as long as the visual context contains the situation arity information of the visual scene.

关键词： vision-language interaction cross-modal reference syntactic disambiguation

Topic-sensitive neural headline generation

在线全文

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2020年第8期63卷 180-195页

作者： Ayana Ziyun WANG Lei XU Zhiyuan LIU maosong sun State Key Laboratory of Intelligent Technology and Systems Tsinghua National Laboratory for Information Science and TechnologyDepartment of Computer Science and TechnologyTsinghua University Department of Computer Information Management Inner Mongolia University of Finance and Economics

Neural models are being widely applied for text summarization, including headline generation,and are typically trained using a set of document-headline pairs. In a large document set, documents can usually be grouped into various topics, and documents within a certain topic may exhibit specific summarization patterns. Most existing neural models, however, have not taken the topic information of documents into consideration. This paper categorizes documents into multiple topics, since documents within the same topic have similar content and share similar summarization patterns. By taking advantage of document topic information, this study proposes a topic-sensitive neural headline generation model(TopicN HG). It is evaluated on a real-world dataset, large scale Chinese short text summarization dataset. Experimental results show that it outperforms several baseline systems on each topic and achieves comparable performance with the state-of-the-art system. This indicates that TopicN HG can generate more accurate headlines guided by document topics.

关键词： neural networks sequence to sequence learning topic LDA headline generation

Portraying User Life Status from Microblogging Posts

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Tsinghua Science and Technology 2013年第2期18卷 182-195页

作者： Jiayu Tang Zhiyuan Liu maosong sun Jiahua Liu the State Key Laboratory of Intelligent Technology and Systems Tsinghua National Laboratory for Information Science and Technology Department of Computer Science and Technology Tsinghua University

Microblogging services nformation and express opinions pro by vide a novel and popular communication scheme for Web users to share publishing short posts, which usually reflect the users＇ daily life. We can thus model the users＇ daily status and interests according to their posts. Because of the high complexity and the large amount of the content of the microblog users＇ posts, it is necessary to provide a quick summary of the users＇ life status, both for personal users and commercial services. It is non-trivial to summarize the life status of microblog users, particularly when the summary is conducted over a long period. In this paper, we present a compact interactive visualization prototype, LifeCircle, as an efficient summary for exploring the long-term life status of microblog users. The radial visualization provides multiple views for a given microblog user, including annual topics, monthly keywords, monthly sentiments, and temporal trends of posts. We tightly integrate interactive visualization with novel and state-of-the-art microblogging analytics to maximize their advantages. We implement LifeCircle on Sina Weibo, the most popular microblogging service in China, and illustrate the effectiveness of our prototype with various case studies. Results show that our prototype makes users nostalgic and makes them reminiscent about past events, which helps them to better understand themselves and others

关键词： text visualization microblogging topic model sentiment analysis keyword extraction