检索结果-南通市图书馆

Optimizing Fine-Tuning in Quantized Language Models:An In-Depth Analysis of Key Variables

Computers, Materials & Continua 2025年第1期82卷 307-325页

作者： Ao Shen Zhiquan Lai Dongsheng Li Xiaoyu Hu National Key laboratory of Parallel and distributed computing National University of Defense TechnologyChangsha410073China Strategic Assessments and Consultation Institute Academy of Military ScienceBeijing100091China

Large-scale Language Models(LLMs)have achieved significant breakthroughs in Natural Language Processing(NLP),driven by the pre-training and fine-tuning *** this approach allows models to specialize in specific tasks with reduced training costs,the substantial memory requirements during fine-tuning present a barrier to broader ***-Efficient Fine-Tuning(PEFT)techniques,such as Low-Rank Adaptation(LoRA),and parameter quantization methods have emerged as solutions to address these challenges by optimizing memory usage and computational *** these,QLoRA,which combines PEFT and quantization,has demonstrated notable success in reducing memory footprints during fine-tuning,prompting the development of various QLoRA *** these advancements,the quantitative impact of key variables on the fine-tuning performance of quantized LLMs remains *** study presents a comprehensive analysis of these key variables,focusing on their influence across different layer types and depths within LLM *** investigation uncovers several critical findings:(1)Larger layers,such as MLP layers,can maintain performance despite reductions in adapter rank,while smaller layers,like self-attention layers,aremore sensitive to such changes;(2)The effectiveness of balancing factors depends more on specific values rather than layer type or depth;(3)In quantization-aware fine-tuning,larger layers can effectively utilize smaller adapters,whereas smaller layers struggle to do *** insights suggest that layer type is a more significant determinant of fine-tuning success than layer depth when optimizing quantized ***,for the same discount of trainable parameters,reducing the trainable parameters in a larger layer is more effective in preserving fine-tuning accuracy than in a smaller *** study provides valuable guidance for more efficient fine-tuning strategies and opens avenues for further research into optimizing LLM

关键词： Large-scale Language Model Parameter-Efficient Fine-Tuning parameter quantization key variable trainable parameters experimental analysis

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

An efficient quantum proactive incremental learning algorithm

引用

Science China(Physics,Mechanics & Astronomy) 2025年第1期Mechanics & Astronomy) .卷 49-57页

作者： Lingxiao Li Jing Li Yanqi Song Sujuan Qin Qiaoyan Wen Fei Gao State Key laboratory of networking and Switching Technology Beijing University of Posts and Telecommunications School of Cyberspace Security Beijing University of Posts and Telecommunications

In scenarios where a large amount of data needs to be learned, incremental learning can make full use of old knowledge, significantly reduce the computational cost of the overall learning process, and maintain high performance. In this paper, taking the MaxCut problem as our example, we introduce the idea of incremental learning into quantum computing, and propose a Quantum Proactive Incremental Learning algorithm(QPIL). Instead of a one-off training of quantum circuit, QPIL contains a multi-phase training on gradually-increased subgraphs of all vertices, proactively reducing large-scale problems to smaller ones to solve in steps, providing an efficient solution for MaxCut. Specifically, some vertices and corresponding edges are randomly selected for training to obtain optimized parameters of the quantum circuit at first. Then, in each incremental phase, the remaining vertices and corresponding edges are gradually added and the parameters obtained from the previous phase are reused in the parameter initialization of the current phase. We perform experiments on 120 different small-scale graphs, and it shows that QPIL performs superior to prevalent quantum and classical baselines in terms of approximation ratio(AR), time cost, anti-forgetting, and solving stability. In particular, QPIL's AR surpasses 20% of mainstream quantum baselines, and the time cost is less than 1/5 of them. The idea of QPIL is expected to inspire efficient and high-quality solutions in large-scale Max Cut and other combinatorial optimization problems.

关键词： variational quantum algorithm incremental learning multi-phase training MaxCut quantum computing

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

distributed optimal consensus control for multiagent systems based on event-triggered and prioritized experience replay strategies

引用

Science China(Information Sciences) 2025年第1期68卷 306-321页

作者： Cuijuan ZHANG Lianghao JI Shasha YANG Xing GUO Huaqing LI Chongqing Key laboratory of Image Cognition Chongqing University of Posts and Telecommunications Key laboratory of Intelligent Perception and computing of Anhui Province Anqing Normal University College of Electronic and Information Engineering Southwest University

This study uses event-triggered(ET) and reinforcement learning methods to investigate the optimal consensus control problem for cooperative-competitive multiagent systems. It proposes a novel distributed ET control strategy, which relies on a prioritized experience replay(PER) policy. This strategy not only conserves communication resources but also ensures acceptable system performance. To implement the proposed method, actor-critic(AC) dual-structured neural networks(NNs) are used to approximate the value function and control policy. In the AC NNs, the weight estimates for the NNs are updated at the moment of event triggering, resulting in a nonperiodic weight adjustment pattern. This approach decreases the computational cost in comparison with the traditional ET mechanism. The PER-based ET mechanism makes full use of valid historical data and effectively establishes a balance between system performance and communication resource ***, it does not require the following two conditions in most existing studies:(1) requirement of the system dynamics model to be known, and(2) persistent excitation. In addition, Zeno behavior is excluded from this study. Finally, a simulation is conducted to confirm the validity of the suggested approach.

关键词： multiagent systems event-triggered mechanism prioritized experience replay dual-structured neural networks optimal consensus control

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Domain generalization with semi-supervised learning for people-centric activity recognition

引用

Science China(Information Sciences) 2025年第1期68卷 171-188页

作者： Jing LIU Wei ZHU Di LI Xing HU Liang SONG Academy for Engineering & Technology Fudan University Shanghai East-bund Research Institute on networking Systems of AI School of Optoelectronic Information and Computer Engineering University of Shanghai for Science & Technology

People-centric activity recognition is one of the most critical technologies in a wide range of real-world applications,including intelligent transportation systems, healthcare services, and brain-computer interfaces. Large-scale data collection and annotation make the application of machine learning algorithms prohibitively expensive when adapting to new tasks. One way of circumventing this limitation is to train the model in a semi-supervised learning manner that utilizes a percentage of unlabeled data to reduce the labeling burden in prediction tasks. Despite their appeal, these models often assume that labeled and unlabeled data come from similar distributions, which leads to the domain shift problem caused by the presence of distribution gaps. To address these limitations, we propose herein a novel method for people-centric activity recognition,called domain generalization with semi-supervised learning(DGSSL), that effectively enhances the representation learning and domain alignment capabilities of a model. We first design a new autoregressive discriminator for adversarial training between unlabeled and labeled source domains, extracting domain-specific features to reduce the distribution gaps. Second, we introduce two reconstruction tasks to capture the task-specific features to avoid losing information related to representation learning while maintaining task-specific consistency. Finally, benefiting from the collaborative optimization of these two tasks, the model can accurately predict both the domain and category labels of the source domains for the classification task. We conduct extensive experiments on three real-world sensing datasets. The experimental results show that DGSSL surpasses the three state-of-the-art methods with better performance and generalization.

关键词： activity recognition deep learning domain generalization semi-supervised learning adversarial training

来源：

同方期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

RepoLike: a multi-feature-based personalized recommendation approach for open-source repositories

引用

Frontiers of Information Technology & Electronic Engineering 2019年第2期20卷 222-237页

作者： Cheng YANG Qiang FAN Tao WANG Gang YIN Xun-hui ZHANG Yue YU Hua-min WANG Key laboratory of Parallel and distributed computing National University of Defense Technology

With the deep integration of software collaborative development and social networking, social coding represents a new style of software production and creation paradigm. Because of their good flexibility and openness,a large number of external contributors have been attracted to the open-source communities. They are playing a significant role in open-source development. However, the open-source development online is a globalized and distributed cooperative work. If left unsupervised, the contribution process may result in inefficiency. It takes contributors a lot of time to find suitable projects or tasks from thousands of open-source projects in the communities to work on. In this paper, we propose a new approach called "RepoLike," to recommend repositories for developers based on linear combination and learning to rank. It uses the project popularity, technical dependencies among projects, and social connections among developers to measure the correlations between a developer and the given projects. Experimental results show that our approach can achieve over 25% of hit ratio when recommending 20 candidates, meaning that it can recommend closely correlated repositories to social developers.

关键词： Social coding Open-source software Personal recommendation GitHub

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

A 2-stage strategy updating rule promotes cooperation in the prisoner's dilemma game

引用

Chinese Physics B 2012年第10期21卷 555-562页

作者：方祥圣朱平刘润然刘恩钰魏贵义 Department of Information Science Anhui Institute of Economic Management networking and distributed computing laboratory Zhejiang Gongshang University Institute for Information Economy Hangzhou Normal University South China University of Technology

In this study,we propose a spatial prisoner＇s dilemma game model with a 2-stage strategy updating rule,and focus on the cooperation behavior of the *** the first stage,i.e.,the pre-learning stage,a focal player decides whether to update his strategy according to the pre-learning factor β and the payoff difference between himself and the average of his *** the player makes up his mind to update,he enters into the second stage,i.e.,the learning stage,and adopts a strategy of a randomly selected neighbor according to the standard Fermi updating *** simulation results show that the cooperation level has a non-trivial dependence on the pre-learning ***,the cooperation frequency decreases as the pre-learning factor increases;but a high cooperation level can be obtained in the intermediate region of- 3〈 β 〈-*** then give some explanations via studying the co-action of pre-learning and *** results may sharpen the understanding of the influence of the strategy updating rule on evolutionary games.

关键词： evolutionary game theory strategy updating social cooperation prisoner's dilemma game

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Social media in Git Hub: the role of @-mention in assisting software development

引用

Science China(Information Sciences) 2017年第3期60卷 50-67页

作者： Yang ZHANG Huaimin WANG Gang YIN Tao WANG Yue YU Key laboratory of Parallel and distributed computing College of ComputerNational University of Defense Technology

Recently, many researches propose that social media tools can promote the collaboration among developers, which are beneficial to the software development. Nevertheless, there is little empirical evidence to confirm that using @-mention has indeed a beneficial impact on the issues in Git Hub. In order to begin investigating such claim, we examine data from two large and successful projects hosted on Git Hub, the Ruby on Rails and the Angular JS. By using qualitative and quantitative analysis, we give an in-depth understanding on how @-mention is used in the issues and the role of @-mention in assisting software development. Our statistical results indicate that, @-mention attracts more participants and tends to be used in the difficult issues. @-mention favors the solving process of issues by enlarging the visibility of issues and facilitating the developers' collaboration. Our study also build an @-network based on the @-mention database we *** the @-network, we investigate its evolution over time and prove that we certainly have the potential to mine the relationships and characteristics of developers by exploiting the knowledge from the @-network.

关键词： issues social media @-mention Git Hub software development

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Ranking Modules for Integrate Testing Based on PageRank Algorithm

引用

Chinese Journal of Electronics 2017年第5期26卷 993-998页

作者： SUN Qi Krishna Ricky MONIZ YUAN Yuyu Key laboratory of Trustworthy distributed computing and Service(BUPT) Ministry of EducationBeijing University of Posts and Telecommunications

The testing industry need to prioritize the limited resources and focus on testing modules whose failure is mostly likely to cause *** paper discusses a method that can rank modules in a software package for integrate testing using the PageR ank *** this algorithm,a sequences of random walks iteratively can find a high likelihood of encountering a node,which is interpreted as it being an important performance *** experiment result prove that the proposed method actually can be used to prioritize testing of specific modules when testing resource are scarce.

关键词： Software testing Integrate testing PageRank algorithm Testing prioritize

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Improved Relevance Ranking in WebGather

引用

Journal of Computer Science & Technology 2001年第5期16卷 410-417页

作者：雷鸣王建勇陈葆珏李晓明 networking & distributed Computer Systems laboratory Department of Computer Science & Technology Peking University Beijing P. R. China

The amount of information on the web is growing rapidly and search engines that rely on keyword matching usually return too many fow quality matches. To improve search results, a challenging task for search engines is how to effectively calculate a relevane ranking for each web page. This paper discusses in what order a search engine should return the URLs it has produced in response to a user's query so as to show more relevant pages first. Emphasis is given on the ranking functions a-dopted by WebGather that take link structure and user popularity factors into account. Bxperimental results are also presented to evaluate the proposed strategy.

关键词： World Wide Web search engine relevance ranking WebGather

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Why API documentation is insufficient for developers: an empirical study

引用

Science China(Information Sciences) 2021年第1期64卷 248-250页

作者： Qiang FAN Yue YU Tao WANG Gang YIN Huaimin WANG Key laboratory of Parallel and distributed computing College of ComputerNational University of Defense Technology

Dear editor,Application programming interface (API) documentation plays an important role in software development and reuse [1] for both API maintainers and API *** documentation helps developers understand and reuse codes effectively [2] and focus their time on desired interfaces and functions instead of the entire system [3].Most high-quality open source projects maintain complete and informative official *** documentation typically conveys detailed specifications,such as class/inter face hierarchies and method descriptions,which can be of great benefit to developers [4].However,despite its authoritativeness and thoroughness,single-sourced official documentation does not always meet the developers'requirements [5].

关键词： API documentation Android online survey Stack Overflow empirical study

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

欢迎您,

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

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

请选择保存的检索档案：

请选择收藏分类：

通借通还

欢迎您,

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

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

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：