检索结果-内蒙古大学图书馆

Partial multi-label learning via label-specific feature corrections

Science China(information Sciences) 2025年第3期68卷 95-109页

作者： Jun-Yi HANG Min-Ling ZHANG School of Computer Science and Engineering Southeast University Key Laboratory of Computer Network and Information Integration(Southeast University) Ministry of Education

Partial multi-label learning（PML） allows learning from rich-semantic objects with inaccurate annotations, where a set of candidate labels are assigned to each training example but only some of them are valid. Existing approaches rely on disambiguation to tackle the PML problem, which aims to correct noisy candidate labels by recovering the ground-truth labeling information ahead of prediction model induction. However, this dominant strategy might be suboptimal as it usually needs extra assumptions that cannot be fully satisfied in real-world scenarios. Instead of label correction, we investigate another strategy to tackle the PML problem, where the potential ambiguity in PML data is eliminated by correcting instance features in a label-specific manner. Accordingly, a simple yet effective approach named PASE, i.e., partial multi-label learning via label-specific feature corrections, is proposed. Under a meta-learning framework, PASElearns to exert label-specific feature corrections so that potential ambiguity specific to each class label can be eliminated and the desired prediction model can be induced on these corrected instance features with the provided candidate labels. Comprehensive experiments on a wide range of synthetic and real-world data sets validate the effectiveness of the proposed approach.

关键词： machine learning multi-label learning partial multi-label learning label-specific features feature correction

来源：评论

学校读者我要写书评

暂无评论

Foundation models for topic modeling: a case study

引用

Frontiers of computer Science 2025年第2期19卷 129-131页

作者： Han ZENG Jia-Ming SUN Chun-Shu LI Zhuying LI Tong WEI School of Computer Science and Engineering Southeast UniversityNanjing 210096China Key Laboratory of Computer Network and Information Integration Southeast UniversityNanjing 210096China

1 Introduction In Natural Language Processing(NLP),topic modeling is a class of methods used to analyze and explore textual corpora,i.e.,to discover the underlying topic structures from text and assign text pieces to different *** NLP,a topic means a set of relevant words appearing together in a particular pattern,representing some specific *** is beneficial for tracking social media trends,constructing knowledge graphs,and analyzing writing *** modeling has always been an area of extensive research in *** methods like Latent Semantic Analysis(LSA)and Latent Dirichlet Allocation(LDA),based on the“bag of words”(BoW)model,often fail to grasp the semantic nuances of the text,making them less effective in contexts involving polysemy or data noise,especially when the amount of data is small.

关键词： words semantic textual

来源：评论

学校读者我要写书评

暂无评论

FHGraph:A Novel Framework for Fake News Detection Using Graph Contrastive Learning and LLM

引用

computers, Materials & Continua 2025年第4期83卷 309-333页

作者： Yuanqing Li Mengyao Dai Sanfeng Zhang School of Cyber Science and Engineering Southeast UniversityNanjing210096China Key Laboratory of Computer Network and Information Integration(Southeast University) Ministry of EducationNanjing210096China

Social media has significantly accelerated the rapid dissemination of information,but it also boosts propagation of fake news,posing serious challenges to public awareness and social *** real-world contexts,the volume of trustable information far exceeds that of rumors,resulting in a class imbalance that leads models to prioritize the majority class during *** focus diminishes the model’s ability to recognize minority class ***,models may experience overfitting when encountering these minority samples,further compromising their generalization *** node-level classification tasks,fake news detection in social networks operates on graph-level samples,where traditional interpolation and oversampling methods struggle to effectively generate high-quality graph-level *** challenge complicates the identification of new instances of false *** address this issue,this paper introduces the FHGraph(Fake News Hunting Graph)framework,which employs a generative data augmentation approach and a latent diffusion model to create graph structures that align with news communication *** the few-sample learning capabilities of large language models(LLMs),the framework generates diverse texts for minority class *** comprises a hierarchical multiview graph contrastive learning module,in which two horizontal views and three vertical levels are utilized for self-supervised learning,resulting in more optimized *** results show that FHGraph significantly outperforms state-of-the-art(SOTA)graph-level class imbalance methods and SOTA graph-level contrastive learning ***,FHGraph has achieved a 2%increase in F1 Micro and a 2.5%increase in F1 Macro in the PHEME dataset,as well as a 3.5%improvement in F1 Micro and a 4.3%improvement in F1 Macro on RumorEval dataset.

关键词： Graph contrastive learning fake news detection data augmentation class imbalance LLM

来源：评论

学校读者我要写书评

暂无评论

Adaptive VDI Session Placement via User Logoff Prediction

引用

Machine Intelligence Research 2025年第1期22卷 189-200页

作者： Wenping Fan Puhui Meng Yu Tian Min-Ling Zhang Yao Zhang School of Computer Science and Engineering Southeast UniversityNanjing210096China VMware Information Technology(China)Limited Beijing100190China Key Laboratory of Computer Network and Information Integration(Southeast University) Ministry of EducationNanjing210096China

After the global pandemic,DaaS(desktop as a service)has become the first choice of many companies’remote working *** the desktops are usually deployed in the public cloud when using DaaS,customers are more cost-sensitive which boosts the requirement of proactive power *** researches in this area focus on virtual desktop infrastructure(VDI)session logon behavior modeling,but for the remote desktop service host(RDSH)-shared desktop pools,logoff optimization is also *** systems place sessions by round-robin or in a pre-defined order without considering their logoff ***,these approaches usually suffer from the situation that few left sessions prevent RDSH servers from being powered-off which introduces cost *** this paper,we propose session placement via adaptive user logoff prediction(SODA),an innovative compound model towards proactive RDSH session ***,an ensemble machine learning model that can predict session logoff time is combined with a statistical session placement bucket model to place RDSH sessions with similar logoff time in a more centralized manner on RDSH ***,the infrastructure cost-saving can be improved by reducing the resource waste introduced by those RDSH hosts with very few hanging sessions left for a long *** on real RDSH pool data demonstrate the effectiveness of the proposed proactive session placement approach against existing static placement techniques.

关键词： Virtual desktop infrastructure(VDI)resource management remote desktop service logoff prediction adaptive modeling session placement

来源：评论

学校读者我要写书评

暂无评论

ADACQR: Enhancing Query Reformulation for Conversational Search via Sparse and Dense Retrieval Alignment 31

ADACQR: Enhancing Query Reformulation for Conversational Sea...

引用

31st International Conference on Computational Linguistics, COLING 2025

作者： Lai, Yilong Wu, Jialong Zhang, Congzhi Sun, Haowen Zhou, Deyu School of Computer Science and Engineering Key Laboratory of Computer Network and Information Integration Ministry of Education Southeast University China

ISBN: (纸本)9798891761964

Conversational Query Reformulation (CQR) has significantly advanced in addressing the challenges of conversational search, particularly those stemming from the latent user intent and the need for historical context. Recent works aimed to boost the performance of CQR through alignment. However, they are designed for one specific retrieval system, which potentially results in sub-optimal generalization. To overcome this limitation, we present a novel framework ADACQR. By aligning reformulation models with both term-based and semantic-based retrieval systems, ADACQR enhances the generalizability of information-seeking queries among diverse retrieval environments through a two-stage training strategy. Moreover, two effective approaches are proposed to obtain superior labels and diverse input candidates, boosting the efficiency and robustness of the framework. Experimental results on the TopiOCQA, QReCC and TREC CAsT datasets demonstrate that ADACQR outperforms the existing methods in a more efficient framework, offering both quantitative and qualitative improvements in conversational query reformulation. © 2025 Association for Computational Linguistics.

关键词： Structured Query Language

来源：评论

学校读者我要写书评

暂无评论

SEED: Accelerating Reasoning Tree Construction via Scheduled Speculative Decoding 31

SEED: Accelerating Reasoning Tree Construction via Scheduled...

引用

31st International Conference on Computational Linguistics, COLING 2025

作者： Wang, Zhenglin Wu, Jialong Lai, Yilong Zhang, Congzhi Zhou, Deyu School of Computer Science and Engineering Key Laboratory of Computer Network and Information Integration Ministry of Education Southeast University China

ISBN: (纸本)9798891761964

Large Language Models (LLMs) demonstrate remarkable emergent abilities across various tasks, yet fall short of complex reasoning and planning tasks. The tree-search-based reasoning methods address this by encouraging the exploration of intermediate steps, surpassing the capabilities of chain-of-thought prompting. However, significant inference latency is introduced due to the systematic exploration and evaluation of multiple thought paths. This paper introduces SEED, a novel and efficient inference framework to improve both runtime speed and GPU memory management concurrently. Based on a scheduled speculative execution, SEED efficiently handles multiple iterations for thought generation and state evaluation, leveraging a rounds-scheduled strategy to manage draft model dispatching. Extensive experimental evaluations on three reasoning datasets demonstrate the superior speedup performance of SEED. © 2025 Association for Computational Linguistics.

关键词： Memory management

来源：评论

学校读者我要写书评

暂无评论

Enhancing Parameter-Efficient Transformers with Contrastive Syntax and Regularized Dropout for Neural Machine Translation 21st

Enhancing Parameter-Efficient Transformers with Contrastive...

引用

21st Pacific Rim International Conference on Artificial Intelligence, PRICAI 2024

作者： Fan, Yunlong Yang, Zhiheng Li, Baixuan Gao, Zhiqiang School of Computer Science and Engineering Southeast University Nanjing China Key Laboratory of Computer Network and Information Integration Southeast University Ministry of Education Nanjing China

ISBN: (纸本)9789819601219

Transformers have significantly improved Neural Machine Translation (NMT) models, accompanied by the inherent space complexity of O(n2). While recent approaches aim to be parameter-efficient, they often exhibit limited generalization capabilities with fewer parameters and decreased performances in longer sentences. To this end, we propose two methods, Syntax-enhanced Contrastive Learning (Syn-CL) and JS divergence-based Regularized Dropout (JSR-Drop) for the cross-layer parameter-sharing Transformer (baseline) in NMT. In Syn-CL, we add the corresponding target-to-source instances for the same batch in an extra bidirectional training phase, and minimize the distances of syntax-enhanced representation between the bilingual sentence-pair while maximizing the ones among others. In JSR-Drop, we refine the regularized dropout strategy with JS divergence, and improve generalization capabilities by reducing Transformers’ inconsistencies between training and inference introduced by dropout. Extensive experiments on six NMT tasks, IWSLT2014 (German↔English) and IWSLT2017 (English↔French and English↔Romanian), show an average BLEU score of 37.55, surpassing the vanilla Transformer and the baseline up to 1.75 BLEU, and outperforming models with 1.5×, 2.0× parameters in translating long sentences. Our code is available at https://***/Anon1214/EnPeNMT. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Contract-Based Incentive Mechanisms for UAV networks in Covert Communications

引用

IEEE Transactions on Vehicular Technology 2025年第5期74卷 7378-7391页

作者： Tian, Wen Fan, Hao Liu, Guangjie Du, Miao Liu, Yinqiu Niyato, Dusit Han, Zhu Nanjing University of Information Science and Technology School of Electronic and Information Engineering Nanjing University of Information Science and Technology Key Laboratory of Intelligent Support Technology In Complex Environment Ministry of Education Nanjing210044 China Southeat University School of Computer Science and Engineering China Southeast University Key Laboratory of Computer Network and Information Integration Ministry of Education Nanjing China Nanyang Technological University School of Computer Science and Engineering 639798 Singapore University of Houston Department of Electrical and Computer Engineering HoustonTX77004 United States Kyung Hee University Department of Computer Science and Engineering Seoul446-701 Korea Republic of

Covert communication is considered a promising technology for hiding transmission processes and activities from malicious eavesdropping. With the development of detection technology, the traditional point-to-point covert communication method cannot effectively guarantee the reliability of covert communication. In order to solve this problem, unmanned aerial vehicle (UAV) are widely adopted in covert communication as common relay equipment. However, the UAV maximizes its benefits by taking advantage of the information asymmetry between itself and the covert communication sender. Therefore, it is necessary to encourage UAV-assisted potential users (UPU) to transmit covert data honestly by designing an information asymmetry-based incentive mechanism. In this paper, we propose a UAV-assisted contract-theoretic (UACT) model to improve the covert communication performance, where UPU honestly transmits covert data and the throughput of UACT can be increased. Specifically, we first divide UPU into finite types and model the data transfer contract between the sender, UPU, and receiver. Then, the contract feasibility of UACT is derived with sufficient and necessary conditions. Finally, we analyze the contract optimization in the continuous case of UPUs. Numerical simulations show that the UACT can incentivize UPUs to transmit covert data through UAV assistance and make covert communication performance close to the information symmetry case. © 1967-2012 IEEE.

关键词： Unmanned aerial vehicles (UAV)

来源：评论

学校读者我要写书评

暂无评论

LASS: A Novel and Economical Data Augmentation Framework Based on Language Models for Debiasing Opinion Summarization 31

LASS: A Novel and Economical Data Augmentation Framework Bas...

引用

31st International Conference on Computational Linguistics, COLING 2025

作者： Zhang, Yanyue Li, Pengfei Lai, Yilong He, Yulan Zhou, Deyu School of Computer Science and Engineering Key Laboratory of Computer Network and Information Integration Ministry of Education Southeast University China Department of Informatics King's College London United Kingdom The Alan Turing Institute United Kingdom

ISBN: (纸本)9798891761964

As more than 70% of reviews in the existing opinion summary data set are positive, current opinion summarization approaches are hesitant to generate negative summaries given the input of negative texts. To address such sentiment bias, a direct approach without the reliance on a specific structure is to generate additional data based on large language models to balance the emotional distribution of the dataset. However, large-scale data augmentation based on large language models faces an apparent disadvantage, the expensive costs. Therefore, in this paper, we propose LASS, a novel data augmentation framework based on both LArge and Small language models for debiaSing opinion summarization. Specifically, a small number of synthesized negative reviews is obtained by rewriting the positive text via a large language model. Then, a disentangle reconstruction model is trained based on the generated data. After training, a large amount of synthetic data can be obtained by decoding the new representation obtained from the combination of different sample representations and filtering based on perplexity degree and sentiment classification. Experiments have proved that LASS can effectively alleviate emotional bias, similar to using only large models, but in a more economical way. © 2025 Association for Computational Linguistics.

关键词： Data assimilation

来源：评论

学校读者我要写书评

暂无评论

HIFI-Stego: A High-Fidelity Embedding Audio Steganography Based on Audio Features Decoupling

IEEE Transactions on Audio, Speech and Language Processing

引用

IEEE Transactions on Audio, Speech and Language Processing 2025年 33卷 2032-2044页

作者： Sanfeng Zhang Baiyu Tian Yang Gao Xingyu Liu Wang Yang School of Cyber Science and Engineering Southeast University Nanjing China Key Laboratory of Computer Network and Information Integration Southeast University Ministry of Education Nanjing China

The higher audio quality of steganography is directly correlated with the increased resistance to steganalysis tools. The advancement of generative AI technologies, particularly those that decouple style features from content, has shown promising developments by facilitating the creation of superior media content. This paper introduces the concept of audio decoupling and presents HIFI-Stego, an embedding audio steganography technique that aims to improve security while maintaining elevated stego audio quality. HIFI-Stego comprises a generator based on the encoder-decoder architecture and a secret message extractor. The encoder of the generator decouples the original audio, yielding the content vector, while the vocoder WORLD is employed to extract the style vector to preserve high-quality audio related features such as timbre and tone. Subsequently, the decoder embeds the secret vector into the decoupled content vector and then couples it with the style vector to generate high-fidelity stego audio. As embedding is not done in traditional time domain or frequency domain, existing analysis tools targeted at traditional steganographies fail to effectively detect the presence of the hidden message. The secret message extractor reuses the generator encoder and augments it with a single-layer convolutional neural network, resulting in a simplified structure suitable for lightweight deployment. Experimental results demonstrate that HIFI-Stego outperforms traditional generative and embedding steganographies in terms of audio quality, steganographic capacity, anti-analysis ability, and concealment.

关键词： Steganography Vectors Time-domain analysis Generators Feature extraction Training Resistance Generative adversarial networks Convolutional neural networks Security

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：