检索结果-内蒙古大学图书馆

24th International Conference on Algorithms and Architectures for Parallel Processing, ICA3PP 2024

作者： Gao, Zhipeng Li, Yichen Yu, Xinlei The State Key Laboratory of Networking and Switching Technology Beijing University of Posts and Communications Beijing China

ISBN: (纸本)9789819615278

Large language models (LLMs) trained on massive publicly accessible data excel in general domain tasks. However, the performance of LLMs in specialized domains such as healthcare and mathematics frequently fails to meet established benchmarks. The primary obstacle to improving this performance is the scarcity of high-quality and domain-specific data due to stringent data privacy regulations. Federated learning, which facilitates the training of LLMs across distributed devices through a decentralized approach that precludes centralized data aggregation presents a potential solution. Nevertheless, federated learning for instruction tuning encounters challenges including data heterogeneity and substantial computational resource requirements. Consequently, this paper proposes FEDNPAIT, an innovative federated learning framework for instruction tuning that incorporates Nesterov-accelerated momentum and a new parameter for controlling the adaptive strength of gradient updates during optimization to improve the performance of LLMs in heterogeneous federated learning environments. Furthermore, FEDNPAIT effectively addresses computational resource limitations through the incorporation of Low-Rank Adaptation. Experimental results indicate that FEDNPAIT effectively mitigates the adverse effects of data heterogeneity in federated environments and outperforms existing federated learning algorithms. Specifically, the results from tuned Llama3-8B demonstrate enhanced generalization and robustness, illustrating how to incorporate complex optimization algorithms into federated training of LLMs, and opening new possibilities for federated instruction tuning in sensitive industries. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Federated learning

来源：评论

学校读者我要写书评

暂无评论

MDR: Model-Specific Demonstration Retrieval at Inference Time for In-Context Learning

MDR: Model-Specific Demonstration Retrieval at Inference Tim...

引用

2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL 2024

作者： Wang, Huazheng Wu, Jinming Sun, Haifeng Xia, Zixuan Cheng, Daixuan Wang, Jingyu Qi, Qi Liao, Jianxin State Key Laboratory of Networking and Switching Technology Beijing University of Posts and Telecommunications China

ISBN: (纸本)9798891761148

Recently, retrieval-based in-context learning (ICL) methods for selecting demonstrations have been widely investigated. Existing methods train a dense retriever to retrieve the most appropriate demonstrations for a given test query, which improves ICL performance. However, we find that distinct LLMs exhibit different biases for "what is a good demonstration" since they possess differences in training data, model architectures and training methods. As a result, a demonstration suitable for one LLM may not be appropriate for others. Previous approaches ignore the model bias and fail to retrieve the most appropriate demonstrations for different inference LLMs, resulting in a degradation of ICL performance. To address this problem, we propose a simple yet effective metric to evaluate the appropriateness of demonstrations for a specific inference LLM. Furthermore, we introduce a Model-specific Demonstration Retrieval (MDR) method for ICL at inference time, which considers the biases of different LLMs. We test MDR on seen and unseen tasks with multi-scale inference LLMs, such as GPT-Neo-2.7B, LLaMA-7B and Vicuna-13B. Experiments on 23 datasets across 11 data domains highlight the remarkable effectiveness of MDR, showcasing improvements of up to 41.2% in comparison to methods that neglect model biases. Our code will be publicly available at: https://***/kiming-ng/MDR. ©2024 Association for Computational Linguistics.

关键词： Demonstrations

来源：评论

学校读者我要写书评

暂无评论

Resource Awareness and Dynamic Load Balancing in Satellite-Ground Integrated Networks 4

Resource Awareness and Dynamic Load Balancing in Satellite-G...

引用

4th International Conference on Artificial Intelligence and Electromechanical Automation, AIEA 2023

作者： Ji, Xiaoyuan Yao, Haipeng Gong, Yongkang State Key Laboratory of Networking and Switching Technology BUPT China

ISBN: (纸本)9781510666412

With the rapid development of mobile communications, many applications have covered industry, healthcare, education, transportation, etc. The traditional ground network has limited coverage range and service capacity, which makes it hard to meet the transmission requirements of the above applications. Satellite-ground integrated networks (SGINs) are introduced to improve data rate and enlarge the network capacity. However, the management strategy of communication resources among these heterogeneous nodes differs greatly. Software defined networking (SDN) simplifies network management via decoupling forward plane and control plane, which is beneficial to boosting the central management for SGINs. Based on SDN architecture, we propose perceptual load balancing to make forwarding decisions. Furthermore, we downscale the substrate network to reduce the communication overhead. Massive experiments validate that our proposed multiple resources-aware based dynamic load balancing strategy extends the average network service time for about 24%, while the network overhead of transmitting information is reduced by 80%. © 2023 SPIE.

关键词： Satellites

来源：评论

学校读者我要写书评

暂无评论

A Low-Ops IaaS Cloud Framework based on Cloud-native Architecture 24

A Low-Ops IaaS Cloud Framework based on Cloud-native Archite...

引用

7th International Conference on Software Engineering and Information Management, ICSIM 2024

作者： Zhao, Haotian Sun, Yuchen State Key Laboratory of Networking and Switching Technology Beijing University of Posts and Telecommunications China

ISBN: (纸本)9798400709197

With the development of Infrastructure as a Service (IaaS) systems, the complexities of deployment and configuration processes, alongside the elevated operational and maintenance costs, present numerous challenges. Therefore, this paper proposes a Low-Ops Framework based on Cloud-native Architecture. In this framework, we design a compact IaaS Virtualization Control Access (VCA) Module to simplify the deployment and configuration process. Simultaneously, we develop an efficient Operations and Maintenance (O&M) Module to enhance operational and maintenance efficiency. Additionally, within the VCA Module, a privileged agent is proposed to control VMs on hosts across the container isolation barrier. Finally, we implement a system based on the proposed framework, and extensive experiments demonstrate its effectiveness, offering a 2.3% improvement compared to the state-of-the-art designs. © 2024 ACM.

关键词： Access control

来源：评论

学校读者我要写书评

暂无评论

IMAP-GCG: Edge Container Resource Scheduling and Configuration Method Based on Improved MAPPO and GCN-GRU 13th

IMAP-GCG: Edge Container Resource Scheduling and Configurati...

引用

13th International Conference on Computer Engineering and Networks, CENet 2023

作者： Gong, Xingle Yang, Yang Sun, Yating Gao, Zhipeng Rui, Lanlan State Key Laboratory of Networking and Switching Technology Beijing University of Posts and Telecommunications Beijing China

ISBN: (纸本)9789819992386

Container technology has been widely used in industrial development, including system deployment and program management, and has achieved very good results. However, in the current edge network environment, there are delay problems caused by container cold start and user terminal task changes, which need to be solved by designing a flexible and efficient edge container resource configuration method. In response to this problem, this paper proposes a method for dynamic container resource allocation called Improved Multi-Agent PPO with GCN-GRU (IMAP-GCG). This method mainly implements the prediction of terminal workload, container resource pre-configuration and container resource scheduling methods. First, we propose a timing prediction model with GCN and GRU as the encoder and LSTM as the decoder to predict the temporal periodicity of user requests, so as to deploy containers in advance. Then, a MAPPO algorithm based on trusted region policy optimization is proposed to schedule container resources, which improves the performance of container resource scheduling. Experiments on two workloads show that the IMAP-GCG model proposed in this paper is superior to the existing advanced multi-agent reinforcement learning model in terms of CPU utilization, bandwidth utilization, and task average delay. The task type forecasting accuracy on this indicator is due to the existing advanced time series forecasting models. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd 2024.

关键词： Containers

来源：评论

学校读者我要写书评

暂无评论

RoboFormer: A Robust Multi-Modal Transformer for 3D Object Detection in Autonomous Driving 6

RoboFormer: A Robust Multi-Modal Transformer for 3D Object D...

引用

6th ACM International Conference on Multimedia in Asia, MMAsia 2024

作者： Liu, Yuang Liao, Dacheng Qi, Mengshi Liu, Liang Ma, Huadong State Key Laboratory of Networking and Switching Technology Beijing University of Posts and Telecommunications Beijing China

ISBN: (纸本)9798400712739

Multimodal fusion-based perception models are essential for 3D object detection in autonomous driving. However, existing approaches often lack robustness evaluations and corresponding design considerations for real-world challenges, such as adverse weather conditions and sensor noise. Existing robustness solutions are costly for already-trained models because they either involve fine-tuning on corrupted data or adding extra sensor inputs. In this work, we propose RoboFormer, a model with inherent robustness and strong scalability across multiple modalities. Our approach combines Top-Down and Bottom-Up methods for mapping modal space to BEV, enhancing robustness. We also use a Transformer-based network for multimodal and temporal fusion, improving performance and facilitating diverse modality extension. To further improve robustness, we introduce a Batch Normalization (BN) fine-tuning method that enables models to effectively handle abrupt domain shifts during inference in extreme scenarios, eliminating the need for retraining on corrupted datasets. Experimental results demonstrate that RoboFormer achieves state-of-the-art performance, while the BN fine-tuning method enhances robustness in the presence of data corruption. The source code is available at the project page. © 2024 Copyright held by the owner/author(s). Publication rights licensed to ACM.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Proactive Platooning Based on C-V2X to Relieve Congestion at a Signalized Intersection

引用

China Communications 2023年第2期20卷 155-167页

作者： Wei Gao Yan Shi Shanzhi Chen The State Key Laboratory of Networking and Switching Technology Beijing University of Posts and TelecommunicationsBeijing 100876China The State Key Laboratory of Wireless Mobile Communications China Academy of Telecommunications TechnologyBeijing 100083China

Nowadays urban traffic congestion becomes a major issue due to the increase in vehicles that lead to more commuting time,accidents,traffic violations,and high fuel *** is a good way to solve traffic congestion because it drives with a smaller inter-vehicle distance than *** paper proposes proactive platooning based on C-V2X(cellular vehicle-to-everything)to relieve *** platooning reduces the inter-vehicle distance and increases the throughput of signalized intersections through integrating the networked control platooning and computation-centralized *** and simulation experiments of proactive platooning were conducted to verify the impact of inter-vehicle distance on platooning latency,platooning safety,and traffic *** results show that the optimal inter-vehicle distance under proactive platooning is less than half of human-driving at a signalized intersection.

关键词： platooning communication platooning safety C-V2X signalized intersection

来源：评论

学校读者我要写书评

暂无评论

Graph Neural Network Enhanced Dynamic Routing for UAV-Assisted Delay Tolerant Network 24

Graph Neural Network Enhanced Dynamic Routing for UAV-Assist...

引用

12th International Conference on Communications and Broadband networking, ICCBN 2024

作者： Gao, Ran Tan, Can Fang, Honglin Yu, Peng State Key Laboratory of Networking and Switching Technology Beijing University of Posts and Telecommunications Beijing China

ISBN: (纸本)9798400717109

In certain extreme environments where infrastructure such as base stations is absent, connections between nodes can be intermittent. Traditional networks struggle to function in these conditions. Delay Tolerant Network (DTN) are a network architecture designed for such scenarios. Utilizing Unmanned Aerial Vehicle (UAV) as nodes within DTN can enhance the network's transmission capabilities and compensate for the deficiencies of ground nodes. However, previous routing methods have shortcomings. Given the high speed and greater freedom of movement of UAVs compared to ground nodes, the network topology is subject to constant and significant change. Traditional routing methods, which typically rely on predetermined rules for selecting forwarding nodes, struggle to adapt to such variations. This paper introduces Graph Neural Network (GNN) into routing decisions for DTN to learn network structural information and assist with routing decisions. In this paper, we first model the routing problem within DTN, accounting for the dynamic changes in network topology and limited resources. Subsequently, we propose a GNN-based dynamic routing method. This method utilizes a graph neural network model that takes as input the feature s of nodes at the current moment and the network's connectivity information to generate optimal forwarding nodes for each data packet. The proposed method has been implemented in the Opportunistic networking Environment (ONE) simulator. Experimental results indicate that our method achieves higher delivery ratio, lower latency, and reduced overhead compared to traditional methods. © 2024 ACM.

关键词： Delay tolerant networks

来源：评论

学校读者我要写书评

暂无评论

CSPUMS: Pioneering Integrated Monitoring in Multi-Service Provider Ecosystems 9

CSPUMS: Pioneering Integrated Monitoring in Multi-Service Pr...

引用

9th International Conference on Cloud Computing and Big Data Analytics, ICCCBDA 2024

作者： Liu, Jiali Qin, Yuqin Beijing University of Posts and Telecommunications State Key Laboratory of Networking and Switching Technology Beijing China

ISBN: (纸本)9798350373554

In the era of information interconnection, the evolution of enterprise digitalization necessitates the adoption of architectures that can support dynamic and complex business operations. Microservices architecture, characterized by its flexibility, scalability, and efficiency, has emerged as a pivotal solution for modern enterprises, enabling the decomposition of monolithic applications into independently deployable services. However, with the expansion of enterprises and the intricate nature of business operations, microservices architecture encounters challenges in service governance, monitoring, management, and integration in multi-vendor environments. Addressing these challenges, this paper introduces a Cross-Service Provider Unified Monitoring System (CSPUMS) tailored for microservice business platforms operating within multi-service provider ecosystems. CSPUMS, a novel three-layer system architecture, encompasses a data acquisition layer, a monitoring layer, and a data presentation layer, designed to ensure the reliability, performance, and security of service platforms across diverse cloud environments. Through the deployment of Docker container technology for data aggregation and the implementation of a multi-dimensional threshold alarm template, CSPUMS enhances operational monitoring, fault management, and service continuity in complex multi-service settings. This paper significantly enhances microservice platforms by providing integrated monitoring solutions for multi-cloud and multi-service environments, ensuring operational efficiency and agility for digitally transforming enterprises. © 2024 IEEE.

关键词： Architecture

来源：评论

学校读者我要写书评

暂无评论

LongMuEL: An entity disambiguation method targeting multiple sources of data and long text 7

LongMuEL: An entity disambiguation method targeting multiple...

引用

7th IEEE International Conference on Big Data and Artificial Intelligence, BDAI 2024

作者： Zhang, Erxuan Shuang, Kai Wang, Xianhao Beijing University of Posts and Telecommunications State Key Laboratory of Networking & Switching Technology Beijing China

ISBN: (纸本)9798350352009

Entity disambiguation is an important method for completing knowledge graphs. The data used for knowledge graph completion often integrates various data types and formats. When designing models for entity disambiguation, a key task is to design a knowledge fusion module to efficiently obtain embeddings of candidate entities from the descriptive text and structured data of the candidates. Since the length of candidate entity descriptive text usually far exceeds the limit of input text for pre-trained models, how to incorporate truncated text knowledge into the embeddings of candidate entities is also a problem to be considered when designing entity disambiguation models. In this paper, we propose a novel entity disambiguation model, LongMuEL, which can represent the long descriptive text of candidate entities and integrate structured text from candidate entity knowledge bases. In the long text embedding module, we use multiple attention modules and weight pooling for all truncated text. When obtaining the embedding of structured text, we adopt two self-supervised tasks to pre-train the pre-trained model and generate negative examples through negative sampling. We demonstrate that pre-training the model with two self-supervised tasks results in better embedding effects for structured data. Finally, we fuse the embedding vectors generated from the two data types to calculate the ranking of candidate entities. LongMuEL is a knowledge graph completion model adapted to existing data features, and through ablation experiments, we verify the effectiveness of each part of LongMuEL. © 2024 IEEE.

关键词： Knowledge graph

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：