检索结果-内蒙古大学图书馆

23rd USENIX Conference on File and Storage Technologies, FAST 2025

作者： Chen, Weijian He, Shuibing Qu, Haoyang Zhang, Ruidong Yang, Siling Chen, Ping Zheng, Yi Huai, Baoxing Chen, Gang The State Key Laboratory of Blockchain and Data Security Zhejiang University China Institute of Blockchain and Data Security China Zhejiang Key Laboratory of Big Data Intelligent Computing China Huawei Cloud China

ISBN: (纸本)9781939133458

Modern advanced large language model (LLM) applications often prepend long contexts before user queries to improve model output quality. These contexts frequently repeat, either partially or fully, across multiple queries. Existing systems typically store and reuse the keys and values of these contexts (referred to as prefix KVs) to reduce redundant computation and time to first token (TTFT). When prefix KVs need to be stored on disks due to insufficient CPU memory, reusing them does not always reduce TTFT, as disk I/O latency is high. In this paper, we propose IMPRESS, an importance-informed multi-tier prefix KV storage system to reduce I/O delay for LLM inference by only loading important prefix KVs. IMPRESS first leverages the insight that there is significant similarity in important token index sets across attention heads and introduces an I/O-efficient important KV identification algorithm. It then optimizes prefix KV storage and caching through importance-informed KV management, reducing TTFT during model inference. Our experimental results show that IMPRESS can reduce TTFT by up to 2.8× compared to state-of-the-art systems, while maintaining comparable inference accuracy. © 2025 FAST. All Rights Reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Task Prioritization in Multiagent Environments: A Novel Approach Using Nash Q-Learning

引用

IEEE Transactions on Consumer Electronics 2025年

作者： Li, Kexin Hu, Yu Wang, Jielei Li, Lin Zhang, Shuyuan Luo, Guangchun Laboratory of Intelligent Collaborative Computing University of Electronic Science and Technology of China China Trusted Cloud Computing and Big Data Key Laboratory of Sichuan Province Chengdu611731 China College of Computer Science and Engineering University of Electronic Science and Technology of China Chengdu611731 China

—With the rapid proliferation of smartphones and various terminal devices, edge computing has gained increasing importance in overcoming computational limitations and enhancing service quality through task offloading. However, most existing approaches handle tasks on a first-come, first-served basis, neglecting task prioritization, which is critical in real-world scenarios where tasks have different levels of urgency and resource requirements. This paper addresses the challenge of effective task prioritization in multi-agent task offloading by proposing a Nash Q-learning-based strategy. Our methodology involves optimizing task allocation decisions using Nash Q-learning to integrate Nash equilibrium principles, thereby enhancing adaptability and dynamic decision-making capabilities within a multi-agent environment. The proposed algorithm is evaluated through extensive simulations, comparing total delay, resource utilization, and task success rates across various task offloading strategies. The results demonstrate that our Nash Q-learning approach outperforms conventional techniques by dramatically lowering overall latency, optimizing resource allocation, and maintaining high task success rates. This work demonstrates the efficacy of Nash Q-learning for dynamically setting priorities for tasks in multi-agent edge computing environments, providing a robust framework for effective resource management and collaborative task completion. And also highlight the potential of Nash Q-learning in advancing intelligent task management in edge computing, suggesting its applicability in real-world, resource-constrained environments. © 2025 IEEE. All rights reserved,

关键词： Computation offloading

来源：评论

学校读者我要写书评

暂无评论

Pronounced advance on typhoon track forecast with global convection-permitting model

引用

Science Bulletin 2025年第9期70卷 1424-1428页

作者： Gu, Jun Zhao, Chun Li, Gudongze Feng, Jiawang Xu, Mingyue Du, Qiuyan Xia, Zihan Li, Yubin Chen, Guanghua Hao, Xiaoyu Chen, Junshi An, Hong Deep Space Exploration Laboratory / School of Earth and Space Sciences / CMA-USTC Laboratory of Fengyun Remote Sensing / State Key Laboratory of Fire Science / Institute of Advanced Interdisciplinary Research on High-Performance Computing Systems and Software University of Science and Technology of China Hefei230026 China Laoshan Laboratory Qingdao266237 China CAS Center for Excellence in Comparative Planetology University of Science and Technology of China Hefei230026 China School of Atmospheric Physics Nanjing University of Information Science and Technology Nanjing210044 China Key Laboratory of Cloud-Precipitation Physics and Severe Storms Institute of Atmospheric Physics Chinese Academy of Sciences Beijing100029 China School of Computer Science and Technology University of Science and Technology of China Hefei230026 China

来源：评论

学校读者我要写书评

暂无评论

Towards Robust Video Anomaly Detection in 6G Networks: A Scene-Adaptive Framework

引用

IEEE Network 2025年

作者： Zhang, Sijia Yu, Yong Tao, Xiaoling Wang, Lianhai Shaanxi Normal University School of Computer Science China Guilin University of Electronic Technology Guangxi Cooperative Innovation Center of Cloud Computing and Big Data China Qilu University of Technology Shandong Academy of Sciences Shandong Provincial Key Laboratory of Computing Power Internet and Service Computing China

With the rapid advancement of 5th Generation (5G) networks, their low latency and high bandwidth facilitate the development of smart city systems. Among the smart city systems, anomaly detection in surveillance videos has been an essential part due to its importance to public safety and security. With the transition to 6G technology, the practical applications of video anomaly detection are expanding to encompass a variety of scenarios, such as sidewalks, shopping malls and campuses, which exhibit significant differences. However, most existing anomaly detection methods deployed in the cloud typically rely on the fixed datasets for training and apply the trained uniform model to the edge services across different scenarios. The distinct environmental conditions may lead to varying performance of a uniform anomaly detection model. To alleviate these issues, we propose a scene-adaptive video anomaly detection framework in 6G-based applications. The proposed framework leverages the advantages of 6G networks to facilitate localized processing and rapid adaptation to diverse environments. To enable the anomaly detection model to adapt more robustly to local environments, the proposed framework integrates a pre-trained model and simulates anomalies to train the anomaly detection model in a denoising manner. By integrating pre-trained teacher models with edge computing nodes, the proposed method facilitates accurate anomaly detection while maintaining robust performance across varying scenarios. The framework utilizes both original data and simulated anomalous data to fine-tune the edge model, ensuring it aligns with the unique characteristics of each environment. Experimental results demonstrate the effectiveness of the proposed method across different scenarios, enabling smarter and more responsive surveillance systems in the 6G era. © 2025 IEEE.

关键词： Video analysis

来源：评论

学校读者我要写书评

暂无评论

Removing Regional Steering Vectors to Achieve Knowledge Domain Forgetting in Large Language Models 6th

Removing Regional Steering Vectors to Achieve Knowledge Dom...

引用

6th International Conference on Machine Learning for Cyber Security, ML4CS 2024

作者： Wu, Wei Wang, Chen Xu, Qiuhao Kong, Wei Zhejiang Sci-Tech University Hangzhou310018 China Zhejiang Key Laboratory of Digital Fashion and Data Governance Zhejiang Sci-Tech University Hangzhou310018 China Zhejiang Provincial International Cooperation Base for Science and Technology on Cloud Computing Security and Data Aggregation Zhejiang Sci-Tech University Hangzhou310018 China

ISBN: (纸本)9789819645657

Large language models (LLMs), as a significant direction in the development of artificial intelligence in recent years, are becoming increasingly popular. These models demonstrate significant potential in applications such as personalized services, simulated dialogues, role-playing, and specific compliance requirements. However, with the continuous expansion of LLMs’ practical applications, selective knowledge management of the models—particularly the ability to make the models "forget" certain domain-specific or topic-specific knowledge without impacting their overall performance or knowledge in other areas—has emerged as a critical issue that demands resolution. We proposes an innovative knowledge domain forgetting method designed to enable models to selectively forget specified knowledge without the need for fine-tuning. The method identifies and removes steering vectors associated with specific knowledge domains, thereby achieving effective forgetting of specific knowledge. The proposed approach has been evaluated in several popular open-source LLMs. Experimental results show that this method achieves good knowledge forgetting effects across diverse scenarios and exhibits notable practical value. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Domain Knowledge

来源：评论

学校读者我要写书评

暂无评论

Improving Retrieval-Augmented Deep Assertion Generation via Joint Training

arXiv

引用

arXiv 2025年

作者： Zhang, Quanjun Fang, Chunrong Zheng, Yi Qian, Ruixiang Yu, Shengcheng Zhao, Yuan Zhou, Jianyi Yang, Yun Zheng, Tao Chen, Zhenyu State Key Laboratory for Novel Software Technology Nanjing University China Huawei Cloud Computing Technologies Co. Ltd Department of Computing Technologies Swinburne University of Technology MelbourneVIC3122 Australia

Unit testing attempts to validate the correctness of basic units of the software system under test and has a crucial role in software development and testing. However, testing experts have to spend a huge amount of effort to write unit test cases manually. Very recent work proposes a retrieve-and-edit approach to automatically generate unit test oracles, i.e., assertions. Despite being promising, it is still far from perfect due to some limitations, such as splitting assertion retrieval and generation into two separate components without benefiting each other. In this paper, we propose AG-RAG, a retrieval-augmented automated assertion generation (AG) approach that leverages external codebases and joint training to address various technical limitations of prior work. Inspired by the plastic surgery hypothesis, AG-RAG attempts to combine relevant unit tests and advanced pre-trained language models (PLMs) with retrieval-augmented fine-tuning. The key insight of AG-RAG is to simultaneously optimize the retriever and the generator as a whole pipeline with a joint training strategy, enabling them to learn from each other. Particularly, AG-RAG builds a dense retriever to search for relevant test-assert pairs (TAPs) with semantic matching and a retrieval-augmented generator to synthesize accurate assertions with the focal-test and retrieved TAPs as input. Besides, AG-RAG leverages a code-aware language model CodeT5 as the cornerstone to facilitate both assertion retrieval and generation tasks. Furthermore, AG-RAG designs a joint training strategy that allows the retriever to learn from the feedback provided by the generator. This unified design fully adapts both components specifically for retrieving more useful TAPs, thereby generating accurate assertions. AG-RAG is a generic framework that can be adapted to various off-the-shelf PLMs. We extensively evaluate AG-RAG against six state-of-the-art AG approaches on two benchmarks and three metrics. Experimental results show th

关键词： Software testing

来源：评论

学校读者我要写书评

暂无评论

IMPRESS: an importance-informed multi-tier prefix KV storage system for large language model inference 25

IMPRESS: an importance-informed multi-tier prefix KV storage...

引用

Proceedings of the 23rd USENIX Conference on File and Storage Technologies

作者： Weijian Chen Shuibing He Haoyang Qu Ruidong Zhang Siling Yang Ping Chen Yi Zheng Baoxing Huai Gang Chen The State Key Laboratory of Blockchain and Data Security Zhejiang University and Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security and Zhejiang Key Laboratory of Big Data Intelligent Computing The State Key Laboratory of Blockchain and Data Security Zhejiang University Huawei Cloud

ISBN: (纸本)9781939133458

关键词：

来源：评论

学校读者我要写书评

暂无评论

EagerLog: Active Learning Enhanced Retrieval Augmented Generation for Log-based Anomaly Detection

EagerLog: Active Learning Enhanced Retrieval Augmented Gener...

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025

作者： Duan, Chiming Jia, Tong Yang, Yong Liu, Guiyang Liu, Jinbu Zhang, Huxing Zhou, Qi Li, Ying Huang, Gang School of Software and Microelectronics Peking University Beijing China Institute for Artificial Intelligence Peking University Beijing China National Engineering Research Center for Software Engineering Peking University Beijing China National Key Laboratory of Data Space Technology and System Beijing China Alibaba Cloud Computing Company Hangzhou China

ISBN: (纸本)9798350368741

Logs record essential information about system operations and serve as a critical source for anomaly detection, which has generated growing research interest. Utilizing large language models (LLMs) within a retrieval-augmented generation (RAG) framework for log-based anomaly detection is an effective approach due to its strong generalization capabilities and efficient few-shot performance. However, the effectiveness of this method hinges on the quality of the knowledge source, which can be impacted by noise and changes within the software systems. Facing these problems, in this paper, we propose a novel log-based anomaly detection method named EagerLog, employing active learning to choose the logs for humans to label, thereby adding them to the knowledge source, thus enhancing the knowledge source and maintaining its quality. Our experiments on three open datasets (BGL, Thunderbird, Zookeeper) and one industrial dataset demonstrate that EagerLog can achieve 93.65% F1 score with approximately 10 labeled log sequences, surpassing existing methods by 15.32%. © 2025 IEEE.

关键词： Active Learning Anomaly Detection Log Analysis Retrieval Augmented Generation

来源：评论

学校读者我要写书评

暂无评论

Multi-keyword Searchable Data Auditing for cloud-Based Machine Learning 6th

Multi-keyword Searchable Data Auditing for Cloud-Based Mach...

引用

6th International Conference on Machine Learning for Cyber Security, ML4CS 2024

作者： Yu, Haiyan Ma, Qingru Zhu, Yilu Cui, Yuxin Zhejiang Sci-Tech University Hangzhou310018 China Zhejiang Key Laboratory of Digital Fashion and Data Governance Zhejiang Sci-Tech University Hangzhou310018 China Zhejiang Provincial International Cooperation Base for Science and Technology on Cloud Computing Security and Data Aggregation Zhejiang Sci-Tech University Hangzhou310018 China Zhejiang Sci-Tech University Hangzhou310018 China

ISBN: (纸本)9789819645657

Machine learning (ML) relies on large amounts of data for training, which often requires outsourcing data to cloud servers. This raises concerns about data integrity and security. Preventing data from being tampered with or lost during storage and transmission has become a critical issue in the ML process. This paper presents a novel data auditing scheme for cloud-based ML, which supports multi-keyword search functionality. It allows searching one or more keywords to ensure the integrity of data in model training. To avoid data privacy leakage, the keywords are encrypted during the auditing process. Only the encrypted keywords of the data required for model training need to be provided to the third-party auditor (TPA). Subsequently, TPA can check and ensure the integrity of the relevant data. Through security analysis and performance evaluations, the results indicate that this scheme can effectively and reliably audit outsourced data while ensuring the security and privacy of model training data. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Data integrity

来源：评论

学校读者我要写书评

暂无评论

Outage Evaluation for STAR-RIS Assisted Satellite-UAV-Terrestrial NOMA Networks With Imperfect CSI

引用

IEEE Internet of Things Journal 2025年

作者： Luo, Wenwei Zhang, Jiliang Song, Jize Liu, Xin Xie, Yiyuan Xu, Jiayou Pan, Gaofeng Southwest University Key Laboratory of Networks and Cloud Computing Security of Universities in Chongqing School of Electronic and Information Engineering Chongqing400715 China Shanghai Institute of Satellite Engineering Shanghai201109 China Beijing Institute of Technology School of Cyberspace Security Beijing100081 China

Due to some transmission delays and channel estimation errors, it is always difficult to acquire perfect channel state information (CSI) in practical communications. Therefore, in this study, taking the practical scenario of imperfect CSI into consideration, a simultaneously transmitting and reflecting-reconfigurable intelligent surface (STAR-RIS)-assisted satellite-unmanned aerial vehicle (UAV)-terrestrial non-orthogonal multiple access (NOMA) network is investigated, in which the UAV is assumed to be randomly located in a spherical cap space while the users are randomly located in an inner circular and an outer annular plane on the ground. Specifically, a satellite source first sends NOMA signals to a UAV relay, which then forwards them to both ground users via a STAR-RIS through reflection as well as transmission. In addition, considering that satellite-UAV link is subject to shadowed-Rician while other links are subject to Nakagami-m distribution, the cumulative distribution functions and probability density functions of channel gains in the presence of imperfect CSI are derived. Furthermore, in practical application scenarios, the positions of the UAV and both users may be randomly located in some regions. Therefore, three randomly distributed scenarios are considered: 1) The users are randomly located while the UAV's position is fixed;2) The UAV is randomly located while both users are in fixed locations;and 3) Both are randomly located. Under those three scenarios, both analytical and asymptotic expressions of outage probability (OP) for two users as well as the system OP are derived using the stochastic geometry approach, and their accuracy is confirmed with Monte-Carlo simulations. © 2014 IEEE.

关键词： Distribution functions

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：