检索结果-内蒙古大学图书馆

2024 International conference on Image processing

作者： Wang, Zhenhua Ye, Linwei Wenzhou Univ Coll Comp Sci & Artificial Intelligence Wenzhou Peoples R China

ISBN: (纸本)9798350349405;9798350349399

The objective of referring image segmentation is to extract referred entities from an image using a particular natural language sentence. The main idea for this task is interacting textual and visual features to build multi-modal relationships. The prior state-of-the-art methods mainly focus on local multi-level intermediate feature interaction or global text-to-image alignment, which might result in insufficient interaction for capturing global multi-modal information exchange or fine-grained referred object details, respectively. To overcome this issue, we introduce a referring image segmentation framework with two-stage multi-modal interaction. Specifically, we devise an innovative multi-level cross-modal fusion module to effectively facilitate the interaction of intermediate features of linguistic and visual modalities for fine-grained details of referred objects. Besides, we further align the linguistic and visual information by introducing an elaborate global alignment module for accurately localizing the entire referred objects. The comprehensive experiments conducted on three referring image segmentation datasets illustrate that our proposed two-stage multi-modal interaction framework exhibits a marked superiority over the contemporary state-of-the-art approaches.

关键词： Vision and language Referring Image Segmentation

来源：评论

学校读者我要写书评

暂无评论

A natural language processing-Based Multimodal Deep Learning Approach for News Category Tagging 8th

A Natural Language Processing-Based Multimodal Deep Learning...

引用

8th International conference on Computer Vision and Image processing (CVIP)

作者： Kumar, Bagesh Singh, Alankar Sharma, Vaidik Shivam, Yuvraj Mohan, Krishna Shukla, Prakhar Falor, Tanay Kumar, Abhishek Indian Inst Informat Technol Allahabad Uttar Pradesh India Manipal Univ Jaipur Rajasthan India

ISBN: (纸本)9783031585340;9783031585357

With the rise in the amount of news available today, the need for its classification has emerged. In this paper, we present methods for tagging news categories using different deep learning models along with a comparison of their effects. These models include single-channel CNN model, multichannel CNN model, and multimodal CNN model. This study involves integration of natural language understanding with convolutional methods that understands descriptions, titles, and tags to enhance news ranking. The novel part of this approach is to find out using natural language understanding with the transfer learning from the supplemental external features that are associated with images. The accuracy of the single-channel model was found to be 81.30%, of the multi-channel model was 85.98% and that of the multi-modal model was 85.39%. We have used the N24 news dataset for the validation of the models.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

Stochastic Fine-Tuning of language Models Using Masked Gradients

Stochastic Fine-Tuning of Language Models Using Masked Gradi...

引用

2024 Findings of the Association for Computational Linguistics, EMNLP 2024

作者： Akbar-Tajari, Mohammad Pilehvar, Mohammad Taher Sharif University of Technology Iran Cardiff University United Kingdom Tehran Institute for Advanced Studies Iran

ISBN: (纸本)9798891761681

Large language Models (LLMs) have emerged as the dominant paradigm in natural language processing owing to their remarkable performance across various target tasks. However, naively fine-tuning them for specific downstream tasks often requires updating a vast number of parameters, resulting in high computational costs and overfitting when training data is limited. In this paper, we propose a novel approach, called Stochastic Tuning, that addresses these challenges by selectively updating a small subset of parameters in each step of the tuning process. Our approach is characterized by its customization of updates based on task-specific partial gradients with respect to stochastic sub-networks. The advantage of Stochastic Tuning over existing solutions lies in its ability to consider both parameter weights as well as forward values which guarantees a context-sensitive fine-tuning. Our experiments demonstrate that Stochastic Tuning outperforms existing lightweight fine-tuning methods, improving average performance by over two points on RoBERTa across several tasks in the GLUE benchmark while updating merely 0.08% of the model's parameters. The code for our implementation can be found at https://***/m-Tajari/StocTuning_LLMs. © 2024 Association for Computational Linguistics.

关键词： Stochastic systems

来源：评论

学校读者我要写书评

暂无评论

Binding language in Administrative Guidance Documents 23

Binding Language in Administrative Guidance Documents

引用

19th International conference on Artificial Intelligence and Law (ICAIL)

作者： Haim, Amit Stanford Law Sch Stanford CA 94305 USA

ISBN: (纸本)9798400701979

Do regulatory guidance documents use binding language despite being purportedly non-binding? Regulatory agencies play a crucial role in modern societies by issuing regulations. While most regulations are promulgated as rules with public notice and comment procedures, administrative guidance documents are as abundant but less studied. They have less formal requirements and are meant as non-binding guidelines, yet skeptics argue they are often used to evade judicial review, and courts turn to their text to inquire whether they are effectively binding. Recent advancements in text analysis methods have allowed scholars to analyze regulatory text, including the measurement of binding language. However, guidance documents have not been part of this trend, largely due to their inaccessibility. This article contributes to the field of empirical legal studies and administrative law by constructing a novel dataset of guidance documents, leveraging a unique policy change. It uses text analysis methods with qualitative insights from doctrinal court decisions, and finds that guidance documents are in fact less binding than rules, but that binding language increased over time and that substantial portions of available documents score higher than a document struck down by a court.

关键词： natural language processing administrative law guidance rules regulation

来源：评论

学校读者我要写书评

暂无评论

A Psycholinguistic Evaluation of language Models' Sensitivity to Argument Roles

A Psycholinguistic Evaluation of Language Models' Sensitivit...

引用

2024 Findings of the Association for Computational Linguistics, EMNLP 2024

作者： Lee, Eun-Kyoung Rosa Nair, Sathvik Feldman, Naomi H. Department of Linguistics University of Maryland College Park United States Institute for Advanced Computer Studies University of Maryland College Park United States

ISBN: (纸本)9798891761681

We present a systematic evaluation of large language models' sensitivity to argument roles, i.e., who did what to whom, by replicating psycholinguistic studies on human argument role processing. In three experiments, we find that language models are able to distinguish verbs that appear in plausible and implausible contexts, where plausibility is determined through the relation between the verb and its preceding arguments. However, none of the models capture the same selective patterns that human comprehenders exhibit during real-time verb prediction. This indicates that language models' capacity to detect verb plausibility does not arise from the same mechanism that underlies human real-time sentence processing. © 2024 Association for Computational Linguistics.

关键词： Computational linguistics

来源：评论

学校读者我要写书评

暂无评论

Exploring the Landscape of natural language processing for Text Analytics: A comprehensive Review 6

Exploring the Landscape of Natural Language Processing for T...

引用

6th International conference on Futuristic Trends in Networks and Computing Technologies, FTNCT 2024

作者： Suthar, Om Prakash Mishra, Ankita Singhal, Shilpa Marwadi Univrsity Gujarat Rajkot36003 India Banasthali Vidyapith Rajasthan Jaipur304022 India

The amount of textual information that can be analyzed in order to look for meaningful information has become a constraint as the amount of digital content that is being produced everyday increases. When it comes to mining large text datasets for useful information, NLP methods and models are necessary and extremely effective tools. The paper's primary objective is to present a comprehensive review of the NLP methods and models that are utilized for text analytics, sentiment analysis, topic modelling, text summarization, and text generation. In this paper, we will discuss the trending methodologies for social media analysis, consumer opinion analysis, and content creation. In addition, we will discuss the methodologies, methods, and evaluation metrics that are utilized in these types of contexts. This analysis aims to provide context for the development of natural language processing (NLP) in the context of text analytics, both historically and prospectively. This will be accomplished by providing context for the development of natural language processing in the context of text analytics. © 2025 The Author(s). Published by Elsevier B.V.

关键词： Sentiment analysis

来源：评论

学校读者我要写书评

暂无评论

DECOUPLING AND REFILLING: A SIMPLE DATA AUGMENTATION METHOD FOR ASPECT TERM EXTRACTION 49

DECOUPLING AND REFILLING: A SIMPLE DATA AUGMENTATION METHOD ...

引用

49th IEEE International conference on Acoustics, Speech, and Signal processing (ICASSP)

作者： Chen, Jiaxiang Hong, Yu Liu, Chaoqun Xu, Qingting Zhou, Guodong Soochow Univ Sch Comp Sci & Technol Suzhou Peoples R China

ISBN: (纸本)9798350344868;9798350344851

Aspect term extraction (ATE) is an important natural language processing task, which aims to extract aspect terms from reviews. Recently, data augmentation has emerged as a reliable approach for relieving data sparsity in the NLP area. For ATE, self-labeling and semi-generation methods have been proposed to implement effective data augmentation. However, they either rely on external data or a pretrained generation model. In this paper, we propose a simple and self-contained augmentation method, which produces new instances for augmentation by context decoupling and infrequent term refilling, without using external data and generation models. We conduct experiments on four benchmark SemEval datasets. The test results show that our method yields substantial improvements, and performs comparably to the state-of-the-art method which uses external data.

关键词： Aspect Term Extraction Data Augmentation Sequence Labeling natural language processing

来源：评论

学校读者我要写书评

暂无评论

AQLoRA: An Adaptive Quantization-Based Efficient Fine-Tuning Method for LLMs 13th

AQLoRA: An Adaptive Quantization-Based Efficient Fine-Tuning...

引用

13th International conference on natural language processing and Chinese Computing

作者： Huang, Xingchen Hu, Yujia Wong, Derek F. Wang, Yao Cal, Liqiong Jiang, Yonghong Guizhou Minzu Univ Sch Data Sci & Informat Engn Guiyang Peoples R China Univ Macau NLP2CT Lab Macau Peoples R China Univ Macau Dept Comp & Informat Sci Macau Peoples R China Guizhou SiSo Elect Co LTD Guiyang Peoples R China

ISBN: (纸本)9789819794331;9789819794348

Large language models (LLMs) have shown exceptional performance in the domain of composite artificial intelligence tasks, offering a preliminary insight into the potential of general artificial intelligence. The fine-tuning process for LLMs necessitates significant computational resources, often surpassing those available from standard consumer-grade GPUs. To this end, we introduce the Adaptive Quantization Low-Rank Adaptation fine-tuning (AQLoRA), a method that reduces memory demands during fine-tuning by utilizing quantization coupled with pruning techniques. This dual strategy not only reduces memory usage but also preserves accuracy. AQLoRA refines the original Low-Rank Adaptation fine-tuning (LoRA) method by efficiently quantizing LLMs weights, prioritizing computational resource allocation based on weight importance, and effectively integrating the quantized model with auxiliary weights post fine-tuning. Applying AQLoRA to the ChatGLM2-6B model, we demonstrate its effectiveness in both natural language generation (NLG) and natural language understanding (NLU) across diverse fine-tuning datasets and scenarios. Our findings reveal that AQLoRA achieves balance between performance and memory efficiency, reducing memory consumption by 25% in NLG tasks. For NLU tasks, it enhances performance by 10% and reduces memory consumption by 10% compared to state-of-the-art methods.

关键词： Large language Model Fine-tuning LoRA

来源：评论

学校读者我要写书评

暂无评论

IFusionQuad: A novel framework for improved aspect-based sentiment quadruple analysis in dialogue contexts with advanced feature integration and contextual CloBlock

引用

EXPERT SYSTEMS WITH APPLICATIONS 2025年 261卷

作者： Jiang, Haoyu Chen, Xiaoliang Miao, Duoqian Zhang, Hongyun Qin, Xiaolin Gu, Xu Lu, Peng Xihua Univ Sch Comp & Software Engn Chengdu 610039 Peoples R China Tongji Univ Coll Elect & Informat Engn Shanghai 201804 Peoples R China Chinese Acad Sci Chengdu Inst Comp Applicat Chengdu 610041 Peoples R China Univ Montreal Dept Comp Sci & Operat Res Montreal PQ H3C 3J7 Canada

Aspect-based sentiment analysis (ABSA) represents a crucial field of natural language processing (NLP). It focuses on deriving detailed sentiment insights from textual content. Dialogue-level aspect-based sentiment quadruple extraction (DiaASQ) is specifically concerned with pinpointing target-aspect-opinion-emotion quadruples within conversations. DiaASQ is important in industries like e-commerce, social media analytics, and customer feedback. However, Current ABSA approaches predominantly focus on single-text scenarios, often overlooking the complexities involved in sentiment analysis within conversational contexts. To fill this gap, this paper presents the IFusionQuad model, which is specifically designed for the DiaASQ task. Our contributions include the innovative integration of CloBlock in ABSA, enhancing feature representation with context-aware weights. The InteractiveNet Fusion Module further advances dialogue understanding by aggregating dialogue- specific features such as threads, speakers, and replies. Components such as CloBlock, gating mechanism, and Biaffine attention effectively mitigate data noise issues, improving the relevance of feature extraction. empirical evaluation on standard datasets demonstrates that the IFusionQuad model outperforms baseline methods, achieving substantial improvements in quadruple extraction. Specifically, our model shows a 6.59% increase in micro F1 and a 7.05% increase in identification F1 for Chinese datasets, and a 2.65% and 4.69% increase in micro F1 and identification F1, respectively, for English datasets. The results clearly demonstrate our IFusionQuad model's efficacy, which consistently outperforms baseline models across all evaluation datasets on the DiaASQ task.

关键词： natural language processing Aspect-based sentiment analysis Aspect sentiment quadruple extraction DiaASQ

来源：评论

学校读者我要写书评

暂无评论

LEARNING AUDIO CONCEPTS FROM COUNTERFACTUAL natural language 49

LEARNING AUDIO CONCEPTS FROM COUNTERFACTUAL NATURAL LANGUAGE

引用

49th IEEE International conference on Acoustics, Speech, and Signal processing (ICASSP)

作者： Vosoughi, Ali Bondi, Luca Wu, Ho-Hsiang Xu, Chenliang Univ Rochester Rochester NY 14627 USA Bosch Res Bosch Ctr Artificial Intelligence Farmington Hills MI USA

ISBN: (纸本)9798350344868;9798350344851

Conventional audio classification relied on predefined classes, lacking the ability to learn from free-form text. Recent methods unlock learning joint audio-text embeddings from raw audio-text pairs describing audio in natural language. Despite recent advancements, there is little exploration of systematic methods to train models for recognizing sound events and sources in alternative scenarios, such as distinguishing fireworks from gunshots at outdoor events in similar situations. This study introduces causal reasoning and counterfactual analysis in the audio domain. We use counterfactual instances and include them in our model across different aspects. Our model considers acoustic characteristics and sound source information from human-annotated reference texts. To validate the effectiveness of our model, we conducted pre-training utilizing multiple audio captioning datasets. We then evaluate with several common downstream tasks, demonstrating the merits of the proposed method as one of the first works leveraging counterfactual information in audio domain. Specifically, the top-1 accuracy in open-ended language-based audio retrieval task increased by more than 43%.

关键词： sound event detection audio understanding multimodal representations free-form text counterfactual representation learning audio captioning

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：