检索结果-内蒙古大学图书馆

arXiv 2024年

作者： He, Yongquan Huang, Xuancheng Tang, Minghao Meng, Lingxun Li, Xiang Lin, Wei Zhang, Wenyuan Gao, Yifu Meituan China Institute of Information Engineering Chinese Academy of Sciences China National Key Laboratory of Parallel and Distributed Computing National University of Defense Technology China

Instruction tuning for large language models (LLMs) can drive them to produce results consistent with human goals in specific downstream tasks. However, the process of continual instruction tuning (CIT) for LLMs may bring about the catastrophic forgetting (CF) problem, where previously learned abilities are degraded. Recent methods try to alleviate the CF problem by modifying models or replaying data, which may only remember the surface-level pattern of instructions and get confused on held-out tasks. In this paper, we propose a novel continual instruction tuning method based on key-part Information Gain (KPIG). Our method computes the information gain on masked parts to dynamically replay data and refine the training objective, which enables LLMs to capture task-aware information relevant to the correct response and alleviate overfitting to general descriptions in instructions. In addition, we propose two metrics, P-score and V-score, to measure the generalization and instruction-following abilities of LLMs. Experiments demonstrate our method achieves superior performance on both seen and held-out tasks. Copyright © 2024, The Authors. All rights reserved.

关键词： Digital storage

来源：评论

学校读者我要写书评

暂无评论

Algorithm for distributed Constraint Optimization Problems with low constraint density

引用

Ruan Jian Xue Bao/Journal of Software 2011年第4期22卷 625-639页

作者： Ding, Bo Wang, Huai-Min Shi, Dian-Xi Tang, Yang-Bin School of Computer National University of Defense Technology Changsha 410073 China National Key Laboratory for Parallel and Distributed Processing School of Computer National University of Defense Technology Changsha 410073 China

Many challenges in multi-agent coordination can be modeled as distributed Constraint Optimization Problems (DCOPs). Aiming at DCOPs with low constraint density, this paper proposes a distributed algorithm based on the idea of greed and backjumping. In this algorithm, each agent makes decisions according to the greedy principle that the most assignment combinations in the problems with low constraint density occur at a zero cost, and the backjumping mechanism among the agents ensures the success of this algorithm, even when this greedy principle leads to a local optimum. In contrast with the existing mainstream DCOP algorithms, this algorithm can solve problems with low constraint density with fewer messages while keeping the polynomial message length and space complexity. The correctness of the key mechanisms in this algorithm has been proved, and those advantages in performance have been verified by experiments. © 2011 ISCAS.

关键词： Multi agent systems

来源：评论

学校读者我要写书评

暂无评论

HyperSpring: Accurate and Stable Latency Estimation in the Hyperbolic Space

HyperSpring: Accurate and Stable Latency Estimation in the H...

引用

International Conference on parallel and distributed Systems (ICPADS)

作者： Yongquan Fu Yijie Wang National Key Laboratory of Parallel and Distributed Processing School of Computer National University of Defense Technology China

Predicting network latencies between Internet hosts can efficiently support large-scale Internet applications, e.g., file sharing service and the overlay construction. Several study use the hyperbolic space to model the Internet dense-core and many-tendril structure. However, existing hyperbolic space based embedding approaches are not designed for accurate latency estimation in the distributed context. We present HyperSpring, which estimates latency by modelling a mass spring system in the hyperbolic similar with Vivaldi. HyperSpring adopts coordinate initialization to speed up the convergence of coordinate computation, uses multiple-round symmetric updates to escape from bad local minima, and stabilizes coordinates by compensating RTT measurements to reduce the coordinate drifts. Evaluation results based on a network trace of 226 PlanetLab nodes indicate that, compared to Euclidean-space based Vivaldi, hyperspring provides performance improvements for most nodes, and incurs slightly higher distortions for a small number of nodes.

关键词： Delay IP networks Web and internet services Large-scale systems Peer to peer computing Springs Convergence Coordinate measuring machines Distortion measurement Extraterrestrial measurements

来源：评论

学校读者我要写书评

暂无评论

Performance problems online detection in cloud computing systems via analyzing request execution paths

Performance problems online detection in cloud computing sys...

引用

International Conference on Dependable Systems and Networks Workshops (DSN-W)

作者： Haibo Mi Huaimin Wang Gang Yin Hua Cai Qi Zhou Tingtao Sun National Laboratory for Parallel and Distributed Processing National University of Defense Technology Changsha China Computing Platform Alibaba Cloud Computing Company Hangzhou China

It is quite a headache for developers to online detect performance problems in large-scale cloud computing systems. The behavior and the hidden connections among the huge amount of runtime request execution paths in cloud computing systems usually contain useful information for performance problem detection. In this paper, we propose an approach to rapidly diagnose the source of performance degradation in large-scale non-stop cloud computing systems. The approach first groups the user requests into categories with a fast clustering algorithm; then applies the principal components analysis to extract the primary methods; finally compares the normal and abnormal behaviors of the primary methods to localize the main cause of performance problems. We conduct extensive experiments over a real-world enterprise system providing services for the public. The results show that our approach can locate the prime causes of performance problems accurately and efficiently.

关键词： Cloud computing Principal component analysis Instruments Degradation Servers Data mining Accuracy

来源：评论

学校读者我要写书评

暂无评论

MatrixMap: Programming Abstraction and Implementation of Matrix Computation for Big Data Applications

MatrixMap: Programming Abstraction and Implementation of Mat...

引用

International Conference on parallel and distributed Systems (ICPADS)

作者： Yaguang Huangfu Jiannong Cao Hongliang Lu Guanqing Liang Department of Computing The Hong Kong Polytechnic University Parallel and Distributed Processing Laboratory National University of Defense Technology

The computation core of many big data applications can be expressed as general matrix computations, including linear algebra operations and irregular matrix operations. However, existing parallel programming systems such as Spark do not have programming abstraction and efficient implementation for general matrix computations. In this paper, we present MatrixMap, a unified and efficient data-parallel system for general matrix computations. MatrixMap provides powerful yet simple abstraction, consisting of a distributed data structure called bulk key matrix and a computation interface defined by matrix patterns. Users can easily load data into bulk key matrices and program algorithms into parallel matrix patterns. MatrixMap outperforms current state-of-the-art systems by employing three key techniques: matrix patterns with lambda functions for irregular and linear algebra matrix operations, asynchronous computation pipeline with optimized data shuffling strategies for specific matrix patterns and in-memory data structure reusing data in iterations. Moreover, it can automatically handle the parallelization and distribute execution of programs on a large cluster. The experiment results show that MatrixMap is 12 times faster than Spark.

关键词： Data structures Programming Linear algebra Machine learning algorithms Sparks Sparse matrices Computational modeling

来源：评论

学校读者我要写书评

暂无评论

Caste-centric agent-oriented programming

Caste-centric agent-oriented programming

引用

International Conference on Quality Software

作者： Ji Wang Rui Shen H. Zhu National Laboratory for Parallel and Distributed Processing Changsha China Department of Computing Oxford Brookes University Oxford UK

The paper presents a caste-centric approach to agent-oriented programming by introducing SLABSp language. The fundamental concepts of caste-centric methodology, caste and scenario, as well as environment descriptions, are available as language facilities in SLABSp in a coherent way. In SLABSp programming, agents are organized into castes to represent their structure and behavior characteristics, and their behaviors are defined by scenarios and rules in the context of their environment. The relations between agents and castes are bound at runtime, and the perceptions and interactions between agents are supported with scenarios and behavior rules. Two selected SLABSp programs are demonstrated to illustrate the programming style.

关键词： Programming profession Computer languages Object oriented modeling Runtime Multiagent systems Research and development Laboratories distributed processing Concurrent computing distributed computing

来源：评论

学校读者我要写书评

暂无评论

Identifying faults in large-scale distributed systems by filtering noisy error logs

Identifying faults in large-scale distributed systems by fil...

引用

International Conference on Dependable Systems and Networks Workshops (DSN-W)

作者： Xiang Rao Huaimin Wang Dianxi Shi Zhenbang Chen Hua Cai Qi Zhou Tingtao Sun National Laboratory for Parallel and Distributed Processing National University of Defense Technology Changsha China Computing Platform Alibaba Cloud Computing Corporation Hangzhou China

Extracting fault features with the error logs of fault injection tests has been widely studied in the area of large scale distributed systems for decades. However, the process of extracting features is severely affected by a large amount of noisy logs. While the existing work tries to solve the problem by compressing logs in temporal and spatial views or removing the semantic redundancy between logs, they fail to consider the co-existence of other noisy faults that generate error logs instead of injected faults, for example, random hardware faults, unexpected bugs of softwares, system configuration faults or the error rank of a log severity. During a fault feature extraction process, those noisy faults generate error logs that are not related to a target fault, and will strongly mislead the resulted fault features. We call an error log that is not related to a target fault a noisy error log. To filter out noisy error logs, we present a similarity-based error log filtering method SBF, which consists of three integrated steps: (1) model error logs into time series and use haar wavelet transform to get the approximate time series; (2) divide the approximate time series into sub time series by valleys; (3) identify noisy error logs by comparing the similarity between the sub time series of target error logs and the template of noisy error logs. We apply our log filtering method in an enterprise cloud system and show its effectiveness. Compared with the existing work, we successfully filter out noisy error logs and increase the precision and the recall rate of fault feature extraction.

关键词： Time series analysis Noise measurement Feature extraction Wavelet transforms Computer crashes Complexity theory Approximation methods

来源：评论

学校读者我要写书评

暂无评论

Fast query over encrypted data based on B+ tree

Fast query over encrypted data based on B+ tree

引用

2009 International Conference on Apperceiving computing and Intelligence Analysis, ICACIA 2009

作者： Wang, Zheng-Fei Tang, Ai-Guo Wang, Wei Department of Computer Hunan Business College Changsha 410205 China National Key Laboratory for Parallel and Distributed Processing School of Computer National University of Defense Technology Changsha 410073 China School of Computer Science Fudan University Shanghai 200433 China

ISBN: (纸本)9781424452040

There are a lot of important and sensitive data in databases, which need to be protected from attacks. To secure the data, Cryptography support is an effective mechanism. However, a tradeoff must be made between the performance and the security because encryption and decryption greatly degrade the query performance. In this paper, a framework is proposed that can quickly implement query over encrypted data based on B+ tree. When data are stored in the form of cipher, it not only stores the encrypted data themselves, but also creates a B+ tree index for the data before encrypting them. When querying the encrypted data, firstly, it locates the encrypted records related to the querying predicate based on the B+ tree index;secondly, it decrypts the encrypted records to accomplish the results. Meanwhile, it must encrypt the B+ tree itself to protect it from leaking confidential information. According to the structure of the B+ tree, it encrypts each node of the B+ tree separately. The results of experiments show that the query performance over the encrypted data decreases about 20 percent compared with the plaintext query performance, and validate the functionality and usability of our approach. ©2009 IEEE.

关键词： Cryptography

来源：评论

学校读者我要写书评

暂无评论

SemanticCast: Content-Based Data Distribution over Self-Organizing Semantic Overlay Networks

SemanticCast: Content-Based Data Distribution over Self-Orga...

引用

IEEE International Conference on parallel and distributed computing, Applications and Technologies (PDCAT)

作者： Zhong Zheng Yijie Wang National Key Laboratory for Parallel and Distributed Processing School of Computer National University of Defense Technology Changsha China

Many applications demand distributing data with different contents efficiently in the network environment with unreliable links and a high node churn. Existing approaches mostly focus on optimizing either efficiency or robustness of data distribution, and fail to ensure both of them simultaneously. In this paper, we propose Semantic Cast - a content-based data distribution approach over self-organizing semantic overlay networks. Semantic Cast maintains a self-organizing semantic overlay based on view exchange (called Crowd). In Crowd, each node seeks neighbors with more similar interests by periodically exchanging its neighbor list (called view) with a chosen neighbor. Through these nodes' self-organizing behavior, various interest communities emerge in the overlay. For data distribution over Crowd, Semantic Cast adopts random walk to route data between interest communities, and adopts flooding to disseminate data inside the interested communities. The experimental results show that compared to existing approaches, Semantic Cast can support efficient content-based data distribution in the unreliable and highly dynamic network environment.

关键词： Semantics Peer to peer computing Communities Routing Robustness Convergence Subscriptions

来源：评论

学校读者我要写书评

暂无评论

NR-MPI: A non-stop and fault resilient MPI supporting programmer defined data backup and restore for E-scale super computing systems

Supercomputing Frontiers and Innovations

引用

Supercomputing Frontiers and Innovations 2016年第1期3卷 4-21页

作者： Suo, Guang Lu, Yutong Liao, Xiangke Xie, Min Cao, Hongjia State Key Laboratory of High Performance Computing National University of Defense Technology Changsha Hunan Province China Science and Technology on Parallel and Distributed Processing Laboratory National University of Defense Technology Changsha Hunan Province China

Fault resilience has became a major issue for HPC systems, particularly, in the perspective of future E-scale systems, which will consist of millions of CPU cores and other components. MPI-level fault tolerant constructs, such as ULFM, are being proposed to support software level fault tolerance. However, there are few systematic evaluations by application programmers using benchmarks or pseudo applications. This paper proposes NR-MPI, a Non-stop and Fault Resilient MPI, supporting programmer defined data backup and restore. To help programmers write fault tolerant programs, NR-MPI provides a set of friendly programming interfaces and a state transition diagram for data backup and restore. This paper focuses on design, implementation and evaluation of NR-MPI. Specifically,this paper puts emphases on failure detection in MPI library, friendly programming interface extending for NR-MPI and examples of fault tolerant programs based NRMPI. Furthermore, to support failure recovery of applications, NR-MPI implements data backup interfaces based on double in-memory checkpoint/*** conduct experiments with both NPB benchmarks and Sweep3D on TH supercomputer in NSCC-TJ. Experimental results show that NR-MPI based fault tolerant programs can recover from failures online without restarting, and the overhead is small even for applications with tens of thousands of cores. © The Authors 2016.

关键词： Message passing

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：