检索结果-内蒙古大学图书馆

Ensemble Knowledge Distillation for Federated Semi-Supervised Image Classification

Tsinghua Science and Technology 2025年第1期30卷 112-123页

作者： Ertong Shang Hui Liu Jingyang Zhang Runqi Zhao Junzhao Du School of Computer Science and Technology Xidian UniversityXi’an 710126China Engineering Research Center of Blockchain Technology Application and Evaluation Ministry of EducationXi’an 710126China Key Laboratory of Smart Human-Computer Interaction and Wearable Technology of Shaanxi Province Xi’an 710126China

Federated learning is an emerging privacy-preserving distributed learning paradigm,in which many clients collaboratively train a shared global model under the orchestration of a remote *** current works on federated learning have focused on fully supervised learning settings,assuming that all the data are annotated with ground-truth ***,this work considers a more realistic and challenging setting,Federated Semi-Supervised Learning(FSSL),where clients have a large amount of unlabeled data and only the server hosts a small number of labeled *** to reasonably utilize the server-side labeled data and the client-side unlabeled data is the core challenge in this *** this paper,we propose a new FSSL algorithm for image classification based on consistency regularization and ensemble knowledge distillation,called *** algorithm uses the global model as the teacher in consistency regularization methods to enhance both the accuracy and stability of client-side unsupervised learning on unlabeled ***,we introduce an additional ensemble knowledge distillation loss to mitigate model overfitting during server-side retraining on labeled *** experiments on several image classification datasets show that our EKDFSSL outperforms current baseline methods.

关键词： federated learning semi-supervised learning federated semi-supervised learning knowledge distillation

来源：评论

学校读者我要写书评

暂无评论

Auto-PIP: Real-time Identification of Critical Performance Inflection Points in Software Stress Testing 35

Auto-PIP: Real-time Identification of Critical Performance I...

引用

35th IEEE International Symposium on Software Reliability Engineering Workshops, ISSREW 2024

作者： Zhang, Shenglin Xiong, Xiao Li, Mengyao Sun, Yongqian Zhao, Yongxin Chen, Xia Deng, Bowen Pei, Dan Nankai University China Haihe Laboratory of Information Technology Application Innovation China Tianjin Key Laboratory of Software Experience and Human Computer Interaction China Huawei Cloud Tsinghua University Beijing National Research Center for Information Science and Technology China

ISBN: (纸本)9798350367041

Conducting stress testing is essential to ensure the stability and performance of software systems post-launch. Among its various aspects, the real-time identification of the "Smooth Load Area"(SLA) is crucial for establishing performance benchmarks, identifying bottlenecks, guiding optimization strategies, and strategically allocating resources. However, existing methods are incapable of real-time inflection point detection or require manual intervention. This paper proposes Auto-PIP, an automated identification framework for performance inflection points based on key performance indicators (KPIs) during performance stress tests. Auto-PIP integrates trend testing algorithms with unsupervised anomaly detection algorithms to identify optimal operating points and estimate maximum capacity. Auto-PIP has been deployed in Huawei Cloud, and the evaluation results show that Auto-PIP demonstrated 100% accuracy in optimal inflection point detection, a 41.7% leap over baselines, and 83.9% accuracy in maximum point identification, a 10.7% gain. Additionally, we have released the dataset to the public to promote ongoing research. © 2024 IEEE.

关键词： Software testing

来源：评论

学校读者我要写书评

暂无评论

Multivariate Time Series Anomaly Detection based on Pre-trained Models with Dual-Attention Mechanism 35

Multivariate Time Series Anomaly Detection based on Pre-trai...

引用

35th IEEE International Symposium on Software Reliability Engineering Workshops, ISSREW 2024

作者： Sun, Yongqian Guo, Yang Liang, Minghan Wen, Xidao Kuang, Junhua Zhang, Shenglin Li, Hongbo Xia, Kaixu Pei, Dan Nankai University China Tianjin Key Laboratory of Software Experience and Human Computer Interaction China BizSeer Haihe Laboratory of Information Technology Application Innovation China Tencent Tsinghua University Beijing National Research Center for Information Science and Technology China

ISBN: (纸本)9798350367041

In major tech companies, monitoring server performance data with anomaly detection algorithms is crucial for assessing operational status. Existing models often require separate training or fine-tuning for each server due to generalization limitations, leading to increased storage, memory, and training costs. As the number of servers grows, this approach becomes impractical. To address this, we propose using pretrained language models for time series anomaly detection, leveraging their strong generalization capabilities. Specifically, we employ two pre-trained GPT-2 models as backbones and implement a two-stage fine-tuning strategy to retain learned knowledge while adapting to specific business data characteristics. Our experiments on multiple anomaly detection datasets demonstrate that our method achieves the best average F1-Score, outperforming the leading baseline by 7%. © 2024 IEEE.

关键词： Time series

来源：评论

学校读者我要写书评

暂无评论

Enhanced Fine-Tuning of Lightweight Domain-Specific Q&A Model Based on Large Language Models 35

Enhanced Fine-Tuning of Lightweight Domain-Specific Q&A Mode...

引用

35th IEEE International Symposium on Software Reliability Engineering Workshops, ISSREW 2024

作者： Zhang, Shenglin Zhu, Pengtian Ma, Minghua Wang, Jiagang Sun, Yongqian Li, Dongwen Wang, Jingyu Guo, Qianying Hua, Xiaolei Zhu, Lin Pei, Dan Nankai University China Haihe Laboratory of Information Technology Application Innovation China Microsoft Tsinghua University China Tianjin Key Laboratory of Software Experience and Human Computer Interaction China China Mobile Research Institute China Beijing National Research Center for Information Science and Technology China

ISBN: (纸本)9798350367041

Large language models (LLMs) excel at general question-answering (Q&A) but often fall short in specialized domains due to a lack of domain-specific knowledge. Commercial companies face the dual challenges of privacy protection and resource constraints when involving LLMs for fine-tuning. This paper propose a novel framework, Self-Evolution, designed to address these issues by leveraging lightweight open-source LLMs through multiple iterative fine-tuning rounds. To enhance the efficiency of iterative fine-tuning, Self-Evolution employ a strategy that filters and reinforces the knowledge with higher value during the iterative process. We employed Self-Evolution on Qwen1.5-7B-Chat using 4,000 documents containing rich domain knowledge from China Mobile, achieving a performance score 174% higher on domain-specific question-answering evaluations than Qwen1.5-7B-Chat and even 22% higher than Qwen1.5-72B-Chat. Self-Evolution has been deployed in China Mobile's daily operation and maintenance for 117 days, and it improves the efficiency of locating alarms, fixing problems, and finding related reports, with an average efficiency improvement of over 18.6%. In addition, we release Self-Evolution framework code in https://***/Zero-Pointer/Self-Evolution. © 2024 IEEE.

关键词： Data mining

来源：评论

学校读者我要写书评

暂无评论

LabelEase: A Semi-Automatic Tool for Efficient and Accurate Trace Labeling in Microservices 35

LabelEase: A Semi-Automatic Tool for Efficient and Accurate ...

引用

35th IEEE International Symposium on Software Reliability Engineering, ISSRE 2024

作者： Zhang, Shenglin Che, Zeyu Pan, Zhongjie Nie, Xiaohui Sun, Yongqian Pan, Lemeng Pei, Dan Nankai University China Haihe Laboratory of Information Technology Application Innovation China Chinese Academy of Sciences Computer Network Information Center China Tianjin Key Laboratory of Software Experience and Human Computer Interaction China Huawei China Tsinghua University Beijing National Research Center for Information Science and Technology China

ISBN: (纸本)9798350353884

Trace data is crucial for system observability and maintainability within microservices architectures, and many operation algorithms depend heavily on trace data, including anomaly detection, root cause analysis, etc. However, the actual performance of these algorithms might be unsatisfactory due to the absence of high-quality labeled datasets for effective training and evaluation. Since billions of traces could be generated daily for large-scale microservices, labeling overhead is the main hurdle to obtaining high-quality trace *** this paper, we propose LabelEase, a novel semi-Automatic trace labeling tool, which uses active learning techniques to achieve efficient and accurate trace labeling. For anomaly trace labeling, LabelEase clusters similar traces with a graph-based trace representation technique and selects a few representative traces for human labeling, avoiding labeling most of the traces. For root cause labeling, LabelEase aggregates the labeled anomalous traces and identifies the service's failures for operators to label. Our systematic experiments on two large-scale datasets show that LabelEase achieves over 0.98 F1-score in anomaly trace labeling and 0.89 precision of failure detection in root cause labeling, LabelEase can reduce operators' labeling overhead by more than 99.9%. To the best of our knowledge, we are the first to propose a semi-Automatic trace labeling tool capable of achieving efficient and accurate trace labeling. © 2024 IEEE.

关键词： Labeled data

来源：评论

学校读者我要写书评

暂无评论

Artificial Intelligence Driven Predictive Analysis of Acoustic and Linguistic Behaviors for ASD Identification

IEEE Transactions on Artificial Intelligence

引用

IEEE Transactions on Artificial Intelligence 2024年第11期5卷 5709-5719页

作者： Ashwini, B. Deeptanshu Gulati, Sheffali Shukla, Jainendra Human-Machine Interaction Laboratory Delhi New Delhi110020 India All India Institute of Medical Sciences Department of Pediatrics Child Neurology Division Centre of Excellence and Advanced Research for Childhood Neurodevelopmental Disorders New Delhi110029 India Department of Computer Science and Engineering Delhi New Delhi110020 India

The identification of autism spectrum disorder (ASD) faces challenges due to the lack of reliable biomarkers and the subjectivity in diagnostic procedures, necessitating improved tools for objectivity and efficiency. Being a key characteristic of autism, language impairments are regarded as potential markers for identifying ASD. However, current research predominantly focuses on analyzing language characteristics in English, overlooking linguistic and contextual specificities in other resource-constrained languages. Motivated by these, we developed an artificial intelligence (AI)-based system to detect ASD, utilizing a range of acoustic and linguistic features extracted from dyadic conversations between a child and their communication partner. Validating our model on 76 English-speaking children [35 ASD and 41 typically developing (TD)] and 33 Hindi-speaking children (15 ASD and 18 TD), our extensive analysis of a diverse and comprehensive set of acoustic and linguistic speech attributes, including lexical, syntactic, semantic, and pragmatic elements revealed reliable speech attributes as predictors of ASD. This comprehensive analysis achieved a remarkable macro F1-score of approximately ∼91.30%. We further addressed the influence of linguistic diversity on speech-based ASD assessment by examining speech behaviors in both English and the low-resource language, Hindi. Specific features such as adverbs and distinct roots contributed significantly to ASD classification in English, while the proportion of unintelligible utterances and adposition use held greater importance in Hindi. This study underscores the reliability of speech-based biomarkers in ASD assessment, emphasizing their effectiveness across diverse linguistic backgrounds and highlighting the need for language-specific research in this domain. © 2024 IEEE.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Biogas Fermentation Monitoring research Based on Borderline-SMOTE Algorithm and Stacking Ensemble Learning 13

Biogas Fermentation Monitoring Research Based on Borderline-...

引用

13th International Conference of Information and Communication Technology, ICTech 2024

作者： Xiu, Pengfei Fu, Lijun Sun, Yukun Wang, Hongliang Li, Xuerao Shenyang Institute of Computing Technology Chinese Academy of Sciences University of Chinese Academy of Sciences Liaoning Province Human-Computer Interaction System Engineering Research Center Based on Digital Twin Shenyang China Shandong University Big Data Technology and Cognitive Intelligence Laboratory University of Chinese Academy of Sciences Shenyang China Shenyang Institute of Computing Technology Chinese Academy of Sciences Liaoning Province Human-Computer Interaction System Engineering Research Center Based on Digital Twin Shenyang China

ISBN: (纸本)9798350376258

Anaerobic fermentation is the most important link to gas production, this paper solves the current problem with the help of the current complete machine learning technology, combining perception and intelligent processing, intelligent processing of information, and derive a decision-making basis for its control or management [1]. Aiming at the current problems of the anaerobic fermentation process in biogas engineering, the system design was carried out, and the anaerobic fermentation process of biogas and its main influencing factors were investigated, and the monitoring parameters of the system were selected to be temperature, PH value, liquid level and redox potential [2]. Previous studies are mainly based on a single prediction model, which leads to poor generalization ability of the model. In this study, unbalanced samples are handled by Borderline-SMOTE algorithm. Based on the Stacking integrated learning approach, a common single prediction model is used to construct a combined prediction model. The accuracy, precision, recall, F1 score and area under the ROC curve are used to evaluate the advantages and disadvantages of the combined prediction model over the single prediction model. ©2024 IEEE.

关键词： Decision making

来源：评论

学校读者我要写书评

暂无评论

Auto-PIP: Real-time Identification of Critical Performance Inflection Points in Software Stress Testing

Auto-PIP: Real-time Identification of Critical Performance I...

引用

IEEE International Conference on Software Reliability Engineering Workshops (ISSRE Wksp)

作者： Shenglin Zhang Xiao Xiong Mengyao Li Yongqian Sun Yongxin Zhao Xia Chen Bowen Deng Dan Pei Nankai University Haihe Laboratory of Information Technology Application Innovation Tianjin Key Laboratory of Software Experience and Human Computer Interaction Huawei Cloud Tsinghua University Beijing National Research Center for Information Science and Technology

ISBN: (数字)9798350367041

ISBN: (纸本)9798350367058

Conducting stress testing is essential to ensure the stability and performance of software systems post-launch. Among its various aspects, the real-time identification of the "Smooth Load Area" (SLA) is crucial for establishing performance benchmarks, identifying bottlenecks, guiding optimization strategies, and strategically allocating resources. However, existing methods are incapable of real-time inflection point detection or require manual intervention. This paper proposes Auto-PIP, an automated identification framework for performance inflection points based on key performance indicators (KPIs) during performance stress tests. Auto-PIP integrates trend testing algorithms with unsupervised anomaly detection algorithms to identify optimal operating points and estimate maximum capacity. Auto-PIP has been deployed in Huawei Cloud, and the evaluation results show that Auto-PIP demonstrated 100% accuracy in optimal inflection point detection, a 41.7% leap over baselines, and 83.9% accuracy in maximum point identification, a 10.7% gain. Additionally, we have released the dataset to the public to promote ongoing research.

关键词： Accuracy Manuals Software systems Market research Real-time systems Software reliability Stress Anomaly detection Optimization Testing

来源：评论

学校读者我要写书评

暂无评论

Multivariate Time Series Anomaly Detection based on Pre-trained Models with Dual-Attention Mechanism

Multivariate Time Series Anomaly Detection based on Pre-trai...

引用

IEEE International Conference on Software Reliability Engineering Workshops (ISSRE Wksp)

作者： Yongqian Sun Yang Guo Minghan Liang Xidao Wen Junhua Kuang Shenglin Zhang Hongbo Li Kaixu Xia Dan Pei Nankai University Tianjin Key Laboratory of Software Experience and Human Computer Interaction BizSeer Haihe Laboratory of Information Technology Application Innovation Tencent Beijing National Research Center for Information Science and Technology Tsinghua University

ISBN: (数字)9798350367041

ISBN: (纸本)9798350367058

关键词： Training Adaptation models Time series analysis Production Reliability engineering Data models Software reliability Servers Anomaly detection Monitoring

来源：评论

学校读者我要写书评

暂无评论

Enhanced Fine-Tuning of Lightweight Domain-Specific Q&A Model Based on Large Language Models

Enhanced Fine-Tuning of Lightweight Domain-Specific Q&A Mode...

引用

IEEE International Conference on Software Reliability Engineering Workshops (ISSRE Wksp)

作者： Shenglin Zhang Pengtian Zhu Minghua Ma Jiagang Wang Yongqian Sun Dongwen Li Jingyu Wang Qianying Guo Xiaolei Hua Lin Zhu Dan Pei Nankai University Haihe Laboratory of Information Technology Application Innovation Microsoft Tsinghua University Tianjin Key Laboratory of Software Experience and Human Computer Interaction China Mobile Research Institute Beijing National Research Center for Information Science and Technology

ISBN: (数字)9798350367041

ISBN: (纸本)9798350367058

Large language models (LLMs) excel at general question-answering (Q&A) but often fall short in specialized domains due to a lack of domain-specific knowledge. Commercial companies face the dual challenges of privacy protection and resource constraints when involving LLMs for fine-tuning. This paper propose a novel framework, Self-Evolution, designed to address these issues by leveraging lightweight open-source LLMs through multiple iterative fine-tuning rounds. To enhance the efficiency of iterative fine-tuning, Self-Evolution employ a strategy that filters and reinforces the knowledge with higher value during the iterative process. We employed Self-Evolution on Qwen1.5-7B-Chat using 4,000 documents containing rich domain knowledge from China Mobile, achieving a performance score 174% higher on domain-specific question-answering evaluations than Qwen1.5-7B-Chat and even 22% higher than Qwen1.5-72B-Chat. Self-Evolution has been deployed in China Mobile’s daily operation and maintenance for 117 days, and it improves the efficiency of locating alarms, fixing problems, and finding related reports, with an average efficiency improvement of over 18.6%. In addition, we release Self-Evolution framework code in https://***/Zero-Pointer/Self-Evolution.

关键词： Privacy Filters Large language models Conferences Software reliability Maintenance Iterative methods Data mining Protection Faces

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：