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

Multi-layer multi-pass friction rolling additive manufacturing of Al alloy:Toward complex large-scale high-performance components

引用

International Journal of Minerals,Metallurgy and Materials 2025年第2期32卷 425-438页

作者： Haibin Liu Run Hou Chenghao Wu Ruishan Xie Shujun Chen College of Mechanical&Energy Engineering Beijing University of TechnologyBeijing 100124China Chongqing Research Institute of Beijing university of technology Chongqing 401121China

At present,the emerging solid-phase friction-based additive manufacturing technology,including friction rolling additive man-ufacturing(FRAM),can only manufacture simple single-pass *** this study,multi-layer multi-pass FRAM-deposited alumin-um alloy samples were successfully prepared using a non-shoulder tool *** material flow behavior and microstructure of the over-lapped zone between adjacent layers and passes during multi-layer multi-pass FRAM deposition were studied using the hybrid 6061 and 5052 aluminum *** results showed that a mechanical interlocking structure was formed between the adjacent layers and the adja-cent passes in the overlapped center *** friction and rolling of the tool head led to different degrees of lateral flow and plastic deformation of the materials in the overlapped zone,which made the recrystallization degree in the left and right edge zones of the over-lapped zone the highest,followed by the overlapped center zone and the non-overlapped *** tensile strength of the overlapped zone exceeded 90%of that of the single-pass deposition *** is proved that although there are uneven grooves on the surface of the over-lapping area during multi-layer and multi-pass deposition,they can be filled by the flow of materials during the deposition of the next lay-er,thus ensuring the dense microstructure and excellent mechanical properties of the overlapping *** multi-layer multi-pass FRAM deposition overcomes the limitation of deposition width and lays the foundation for the future deposition of large-scale high-performance components.

关键词： aluminum alloy additive manufacturing solid-state friction stir welding multi-layer multi-pass

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

Exploring Nanoscale Perovskite Materials for Next‑Generation Photodetectors:A Comprehensive Review and Future Directions

引用

Nano-Micro Letters 2025年第2期17卷 46-108页

作者： Xin Li Sikandar Aftab Maria Mukhtar Fahmid Kabir Muhammad Farooq Khan Hosameldin Helmy Hegazy Erdi Akman State Key Laboratory of Pulsed Power Laser technology National University of Defense TechnologyHefei 230037AnhuiPeople’s Republic of China Anhui Laboratory of Advanced Laser technology Hefei 230037AnhuiPeople’s Republic of China Nanhu Laser Laboratory Changsha 410015HunanPeople’s Republic of China Department of Semiconductor Systems Engineering and Clean Energy Sejong UniversitySeoul 05006Republic of Korea Department of Artificial Intelligence and Robotics Sejong UniversitySeoul 05006Republic of Korea School of Engineering Science Simon Fraser UniversityBurnabyBC V5A 1S6Canada Department of Electrical Engineering Sejong UniversitySeoul 05006South Korea Department of Physics Faculty of ScienceKing Khalid UniversityP.O.Box 9004AbhaSaudi Arabia Central Labs King Khalid UniversityAlQura’aP.O.Box 96061413 AbhaSaudi Arabia Scientific and Technological Research and Application Center Karamanoglu Mehmetbey University70100 KaramanTurkey

The rapid advancement of nanotechnology has sparked much interest in applying nanoscale perovskite materials for photodetection *** materials are promising candidates for next-generation photodetectors(PDs)due to their unique optoelectronic properties and flexible synthesis *** review explores the approaches used in the development and use of optoelectronic devices made of different nanoscale perovskite architectures,including quantum dots,nanosheets,nanorods,nanowires,and *** a thorough analysis of recent literature,the review also addresses common issues like the mechanisms underlying the degradation of perovskite PDs and offers perspectives on potential solutions to improve stability and scalability that impede widespread *** addition,it highlights that photodetection encompasses the detection of light fields in dimensions other than light intensity and suggests potential avenues for future research to overcome these obstacles and fully realize the potential of nanoscale perovskite materials in state-of-the-art photodetection *** review provides a comprehensive overview of nanoscale perovskite PDs and guides future research efforts towards improved performance and wider applicability,making it a valuable resource for researchers.

关键词： Nanoscale perovskites Photodetectors Nanosheets Nanorods Nanowires Quantum dots Nanocrystals

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

Underground laboratories · Deep underground observation ·Scientific questions—Insights from observations of multi-physic fields in deep underground labs

引用

Science China Earth Sciences 2025年

作者： Huiqi REN Yun WANG Chang CHEN Guangyu FU Longqing QIU Lianghui GUO Chengliang XIE Yongsheng HE Heping SUN Jiwen TENG Institute of defense Engineering Academy of Military Sciences State Kay Laboratory of Geological Processes and Mineral Resources Key Laboratory of Intraplate Volcanoes and Earthquakes Ministry of Education China University of Geosciences (Beijing) Shanghai Institute of Microsystem and Information technology Chinese Academy of Sciences Innovation Academy for Precision Measurement Science and technology Chinese Academy of Sciences Institute of Geology and Geophysics Chinese Academy of Sciences

This paper presents a comprehensive review on recent development and research conducted in domestic and international underground laboratories. We first introduce the differences in three environments—surface, mountain tunnel cavities and underground coal mine tunnels—by examining cosmic ray background, ambient noises related to gravity and seismic measurement, and electromagnetic noises in magnetic and magnetotelluric measurements. We highlight potential misuse of the term Underground Lab or Deep Underground Lab when describing observations in different physical fields. We introduce unique features of underground coal mine tunnels in China, such as large spaces, ultra-quiet conditions, and ultra-clean environments. When comparing with mountain tunnel cavities and borehole observations, coal mine tunnel observations have superior long-term stability and high precision. Through observations and comparisons of multi-physic fields at surface and the deep underground, we find that the higher SNR seismic observations conducted in deep underground tunnels in coal mines are beneficial to improve velocity tomography of the solid earth. The gravity observation with a Superconducting Quantum Interference Device(SQUID) makes it possibly to capture slow earthquake, which has not been observed previously in the Chinese mainland. SQUID magnetic observations can detect fluctuations as weak as femto-Tesla(fT), enabling us to explore the attenuation of Schumann Resonance down to the solid Earth. This opens opportunities to investigate the connections between the Earth's magnetic field and the interactions within the human brain and heart. To improve the precision of quantum measurement,we should consider the possible effects of weak magnetic disturbances in deep underground environments. Finally, we discuss the importance of deep underground laboratories, observing facilities and techniques deployed in these laboratories, and their possible connection with respect to “deep space” an

关键词：

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

System distortion analysis and compensation of DIFS signals for wideband imaging radar

引用

Science China(Information Sciences) 2015年第2期58卷 42-57页

作者： LIU Yang HOU QingKai XU ShiYou TIAN Biao National university of defense technology

System distortion is inevitable in wideband imaging radar, which needs to be compensated *** traditional wideband imaging radar for space surveillance, a linear frequency modulation signal is transmitted and the echoed signal is received with dechirping processing. In this case, the system distortion changes with the target range, making it hard to compensate the distortion. However, the direct intermediate frequency sampling(DIFS) signal maintains the complete system distortion and avoids the range variant distortion in the dechirped signal. Therefore, it is more convenient to perform system compensation in DIFS signal. In this paper, the distortion factors affecting the wideband radar systems are introduced. Then, the influence of the amplitude phase distortion on the focusing quality is analyzed in detail. Finally, a system compensation method in the frequency domain based on least squares estimation is proposed. In the proposed method, the compensation vectors are extracted from the calibration tower echoes for DIFS compensation. Inverse synthetic aperture radar imagery of targets can be achieved with improved focus quality. Simulations and real-data experiments confirm the effectiveness of the proposal.

关键词： system distortion direct intermediate frequency sampling(DIFS) system compensation least squares estimation(LSE) wideband imaging radar

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

A Multi-instruction Streams Extension Mechanism for SIMD Processor

引用

Chinese Journal of Electronics 2017年第6期26卷 1154-1160页

作者： PENG Yuanxi ZHOU Feng HAI Yue WANG Yaohua National university of defense technology

Multi-media applications contain multibranches loop, it is of low efficiency to map them into traditional Single instruction multiple data(SIMD) structures. Considering the above matter, we proposed a multiinstruction streams extension method for traditional SIMD structures. The main idea is to simultaneously dispatch multiple instruction streams to multiple lanes. Compared with traditional SIMD whose lanes receive the unified single instruction stream but execute conditionally through a lane mask vector, Multi-instruction streams extension grants each of its lanes the ability to receive and execute the instructions of one particular branch path. Thus, it is of high efficiency to map multi-branches loop in applications. The design is finally implemented through Verilog language, and then integrated into the FT-Matrix vector-SIMD chip. Application profiling results shows that the proposed method consumes mere 2.61% area overhead while obtains about 1.8x to 2.4x performance gain.

关键词： Multi-instructions extension Multibranches loop Single instruction multiple data(SIMD) structure

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

Course Drop-out Prediction on MOOC Platform via Clustering and Tensor Completion

引用

Tsinghua Science and technology 2019年第4期24卷 412-422页

作者： Jinzhi Liao Jiuyang Tang Xiang Zhao National university of defense technology

As a supplement to traditional education, online courses offer people, regardless of their age, gender, or profession, the chance to access state-of-the-art knowledge. Nonetheless, despite the large number of students who choose to begin online courses, it is easy to observe that quite a few of them drop out in the middle, and information on this is vital for course organizers to improve their curriculum outlines. In this work, in order to make a precise prediction of the drop-out rate, we propose a combined method MOOP, which consists of a global tensor and local tensor to express all available feature aspects. Specifically, the global tensor structure is proposed to model the data of the online courses, while a local tensor is clustered to capture the inner connection of courses. Consequently, drop-out prediction is achieved by adopting a high-accuracy low-rank tensor completion method, equipped with a pigeon-inspired algorithm to optimize the parameters. The proposed method is empirically evaluated on real-world Massive Open Online Courses(MOOC) data, and is demonstrated to offer remarkable superiority over alternatives in terms of efficiency and accuracy.

关键词： MOOC platform drop-out prediction tensor completion clustering

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

Compressed sensing digital receiver and orthogonal reconstructing algorithm for wideband ISAR radar

引用

Science China(Information Sciences) 2015年第2期58卷 21-30页

作者： HOU QingKai LIU Yang FAN LiJie SU ShaoYing National university of defense technology

This paper proposes a novel design of intermediate frequency(IF) digital receiver for wideband inverse synthetic aperture radar(ISAR) based on compressed sensing(CS). For the convenience in engineering application, we use random sampling in the digital receiver and make it possible to digitize the wideband IF signal using a commercial off-the-shelf analog-to-digital converter with sub-Nyquist sample rate. Besides, a novel basis for the sparse representation of real-valued ISAR radar echoes is built in this paper, and an orthogonal CS reconstructing algorithm is proposed based on this. Using our proposed method, the complex-valued range profile of target can be directly reconstructed from the subsampled real raw echo. The phase information of target range profile, which is very important for the coherent processing in ISAR imaging, is well reserved during the reconstruction. As a result, the down converter and matched filter, which are essential in conventional radar receiver, can be eliminated in our CS digital receiver. A series of simulation validates our design and demonstrates the feasibility of the sub-Nyquist sampling. The simulation results of ISAR imaging verify the validity and superiority of the proposed orthogonal reconstructing method.

关键词： compressed sensing ISAR imaging digital receiver random sampling orthogonal receiver

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

Implementation of ternary Shor's algorithm based on vibrational states of an ion in anharmonic potential

引用

Chinese Physics B 2015年第3期24卷 157-165页

作者：刘威陈书明张见吴春旺吴伟陈平形 College of Computer National University of Defense Technology Science and technology on Parallel and Distributed Processing Laboratory (PDL) National University of Defense Technology College of Science National University of Defense Technology

It is widely believed that Shor＇s factoring algorithm provides a driving force to boost the quantum computing ***, a serious obstacle to its binary implementation is the large number of quantum gates. Non-binary quantum computing is an efficient way to reduce the required number of elemental gates. Here, we propose optimization schemes for Shor＇s algorithm implementation and take a ternary version for factorizing 21 as an example. The optimized factorization is achieved by a two-qutrit quantum circuit, which consists of only two single qutrit gates and one ternary controlled-NOT gate. This two-qutrit quantum circuit is then encoded into the nine lower vibrational states of an ion trapped in a weakly anharmonic potential. Optimal control theory（OCT） is employed to derive the manipulation electric field for transferring the encoded states. The ternary Shor＇s algorithm can be implemented in one single step. Numerical simulation results show that the accuracy of the state transformations is about 0.9919.

关键词： ternary Shor's algorithm anharmonic ion trapping optimal control theory vibrational state

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

Solving multi-scenario cardinality constrained optimization problems via multi-objective evolutionary algorithms

引用

Science China(Information Sciences) 2019年第9期62卷 73-90页

作者： Xing ZHOU Huaimin WANG Wei PENG Bo DING Rui WANG College of Computer National University of Defense Technology College of Systems Engineering National University of Defense Technology

Cardinality constrained optimization problems(CCOPs) are fixed-size subset selection problems with applications in several fields. CCOPs comprising multiple scenarios, such as cardinality values that form an interval, can be defined as multi-scenario CCOPs(MSCCOPs). An MSCCOP is expected to optimize the objective value of each cardinality to support decision-making processes. When the computation is conducted using traditional optimization algorithms, an MSCCOP often requires several passes(algorithmic runs) to obtain all the(near-)optima, where each pass handles a specific cardinality. Such separate passes abandon most of the knowledge(including the potential superior solution structures) learned in one pass that can also be used to improve the results of other passes. By considering this situation, we propose a generic transformation strategy that can be referred to as the Mucard strategy, which converts an MSCCOP into a low-dimensional multi-objective optimization problem(MOP) to simultaneously obtain all the(near-)optima of the MSCCOP in a single algorithmic run. In essence, the Mucard strategy combines separate passes that deal with distinct variable spaces into a single pass, enabling knowledge reuse and knowledge interchange of each cardinality among genetic individuals. The performance of the Mucard strategy was demonstrated using two typical MSCCOPs. For a given number of evolved individuals, the Mucard strategy improved the accuracy of the obtained solutions because of the in-process knowledge than that obtained by untransformed evolutionary algorithms, while reducing the average runtime. Furthermore, the equivalence between the optimal solutions of the transformed MOP and the untransformed MSCCOP can be theoretically proved.

关键词： evolutionary computation multi-objective optimization cardinality-constrained optimization problem multiple scenarios transformation p-median problem portfolio optimization problem

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：