检索结果-内蒙古大学图书馆

arXiv 2025年

作者： Li, Kewei Kong, Yanwen Xu, Yiping Su, Jianlin Huang, Lan Zhang, Ruochi Zhou, Fengfeng College of Computer Science and Technology Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University Jilin Changchun130012 China Moonshot AI Ltd. Beijing100086 China

Since the emergence of research on improving the length extrapolation capabilities of large models in 2021, some studies have made modifications to the scaling factor in the scaled dot-product attention mechanism as part of their proposed methods without rigorous theoretical justifications. To fill this gap, we propose two new scaled temperatures based on entropy invariance to enhance length extrapolation. First, a training-free method InfoScale is designed for dot-product attention, and preserves focus on original tokens during length extrapolation by ensuring consistent entropy. Second, we theoretically analyze the impact of scaling (CosScale) on cosine attention. Experimental data demonstrates that combining InfoScale and CosScale achieves state-of-the-art performance on the GAU-α model with a context window extended to 64 times the training length, and outperforms seven existing methods. Our analysis reveals that significantly increasing CosScale approximates the Windowed Attention, and highlights the significance of attention score dilution as a key challenge in long-range context handling. The code and data are available at https://***/HT-NEKO/InfoScale. © 2025, CC BY-NC-ND.

关键词： Data handling

来源：评论

学校读者我要写书评

暂无评论

An Enhanced Approach for Few-Shot Segmentation Via Smooth Downsampling Mask and Label Smoothing Loss

SSRN

引用

SSRN 2023年

作者： Jin, Hailong Li, Huiying College of Computer Science and Technology Jilin Province Changchun China Key Laboratory of Symbol Computation and Knowledge Engineering Jilin University Changchun130012 China

Few-shot semantic segmentation aims to segment new categories with only a small number of annotated images. Previous methods mainly focused on exploiting the pixel-level correlation between the support image and the query image, combined with attention-based methods, resulting in significant advancements. In this paper, we introduce a new perspective to enhance few-shot segmentation. We identify that utilizing the bilinear interpolation method to downsample the mask leads to the loss of fine-grained information from the target features. To address this issue, we propose a Smooth Downsampling Mask (SDM) method. The SDM method is designed to retain more effective target semantic features by employing a cascaded downsampling approach with a smooth kernel for mask processing. Additionally, we propose a label smoothing loss to further enhance the performance, which provides direct guidance for low-resolution feature map optimization. Notably, our proposed method involves no additional learnable parameters and introduces only a small amount of extra computation. To validate the effectiveness of our approach, we conduct extensive experiments on three public benchmarks, PASCAL-$5^i$, COCO-$20^i$, and FSS-$1000$, and achieve considerable improvements. © 2023, The Authors. All rights reserved.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

TagVibra: push the limit of RFID-based vibration sensing 9

TagVibra: push the limit of RFID-based vibration sensing

引用

9th International Conference on Big Data Computing and Communications, BigCom 2023

作者： Li, Jian Li, Ping Zhu, Biaokai Dong, Chong Gong, Chao Lv, Zhao Anhui University Anhui Province Key Laboratory of Multimodal Cognitive Computation School of Computer Science and Technology Hefei230601 China Anhui University of Technology Anhui Engineering Research Center for Intelligent Applications and Security of Industrial Internet Ma'anshan243032 China Shanxi Police College Taiyuan030401 China

ISBN: (纸本)9798350331240

RFID-based mechanical vibration detection is considered a promising method for many Internet of Things (IoT) applications. However, existing methods are affected by ambient interference and leakage from the reader signature, which makes it difficult to obtain high-precision detection results. To overcome this limitation, this paper introduces TagVibra, a high-precision non-contact vibration sensing system based on RFID technology. By leveraging the nonlinear effect inside the tag and the modulation mechanism of RFID, TagVibra can achieve sub-millimeter level vibration detection accuracy. We implement TagVibra on our Universal Software Radio Peripheral (USRP) platform and evaluate its performance via experiments. The experimental results show that TagVibra can detect vibration with amplitude as low as 0.3mm, and the average estimation error below 0.7Hz when the vibration frequency is 100Hz. © 2023 IEEE.

关键词： Frequency estimation

来源：评论

学校读者我要写书评

暂无评论

AN ADAPTIVE POLICY TO EMPLOY SHARPNESS-AWARE MINIMIZATION

arXiv

引用

arXiv 2023年

作者： Jiang, Weisen Yang, Hansi Zhang, Yu Kwok, James Guangdong Provincial Key Laboratory of Brain-inspired Intelligent Computation Department of Computer Science and Engineering Southern University of Science and Technology China Department of Computer Science and Engineering Hong Kong University of Science and Technology Hong Kong Peng Cheng Laboratory China

Sharpness-aware minimization (SAM), which searches for flat minima by min-max optimization, has been shown to be useful in improving model generalization. However, since each SAM update requires computing two gradients, its computational cost and training time are both doubled compared to standard empirical risk minimization (ERM). Recent state-of-the-arts reduce the fraction of SAM updates and thus accelerate SAM by switching between SAM and ERM updates randomly or periodically. In this paper, we design an adaptive policy to employ SAM based on the loss landscape geometry. Two efficient algorithms, AE-SAM and AE-LookSAM, are proposed. We theoretically show that AE-SAM has the same convergence rate as SAM. Experimental results on various datasets and architectures demonstrate the efficiency and effectiveness of the adaptive policy. © 2023, CC BY.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Few-Shot Fine-Grained Classification of Histological Images

Few-Shot Fine-Grained Classification of Histological Images

引用

Medical Artificial Intelligence (MedAI), IEEE International Conference on

作者： Yingdong Jiang Jing Huang Zihe Jin Leqi Shen Ziyi Zhang Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education College of Software Jilin University Jilin University ChangChun China Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education College of Computer Science and Technology Jilin University Jilin University ChangChun China College of Computer Science and Technology Jilin University ChangChun China

Histological image classification plays a crucial role in cancer diagnosis. However, the acquisition of well-labeled histological images is prohibitively expensive, and obtaining rare abnormal samples is challenging. Therefore, applying few-shot learning methods to histological image classification tasks holds significant clinical value. Nevertheless, existing research predom-inantly relies on coarse-grained image classification approaches based on natural image datasets, which struggle to address the fine-grained challenges encountered in histological image classification, such as intra-class diversity and inter-class similarity. To tackle this issue, this study proposes a novel few-shot fine-grained classification method for histological images, named “Category-Aware Feature Map Reconstruction Network.” This method employs channel weights to localize the differences between inter-class and intra-class regions, composed of intra-class channel weights and inter-class channel weights, collectively referred to as category-aware weights. Specifically, intra-class channel weights indicate the matching degree of salient regions within the support set of a particular class, while inter-class channel weights represent the degree of containing distinct information between classes. The category-aware weights are utilized to transform the support feature maps and query feature maps, generating feature maps that capture differentiating details between categories. Finally, the distance between the transformed query feature map and support feature map is calculated to achieve probabilistic predictions for the categories. On a histological few-shot dataset, this method achieves an accuracy of 90.23% using ResNet-12 as the feature extractor, surpassing the baseline model by 5.24% and outperforming other few-shot methods by at least 10% in the 5-way 10-shot experimental setting. The proposed method exhibits exceptional performance on histological image few-shot datasets, playing a

关键词：

来源：评论

学校读者我要写书评

暂无评论

Short-Term Traffic Flow Prediction Method Based on Spatio-Temporal Characteristics of Complex Road Network

Short-Term Traffic Flow Prediction Method Based on Spatio-Te...

引用

2021 International Conference on Applied Mathematics, Modeling and Computer simulation, AMMCS 2021

作者： Zhu, Huawei Mao, Yihao Zhao, Yiming Shi, Liyuan Shao, Yanli School of Information and Electrical Engineering Zhejiang University City College Hangzhou310015 China Key Laboratory of Complex Systems Modeling and Simulation School of Computer Science and Technology Hangzhou Dianzi University Hangzhou310018 China

ISBN: (纸本)9781643682549

With the development of urbanization, the number of residents' motor vehicles has increased sharply, and traffic congestion problem has become increasingly serious. The construction of Intelligent Traffic System (ITS) has become the main means to alleviate traffic congestion. Short-term traffic flow prediction has guiding significance for residents' travel planning and intelligent management of transportation, and has become one of the research hotspots in intelligent transportation ***, A short-term traffic flow prediction method based on the spatio-temporal characteristics of complex road networks is proposed to further improve the prediction accuracy and reduce the prediction cost. Firstly, a graph convolutional network (GCN) capable of processing non-Euclidean data structures is used to extract the spatial characteristics of traffic flow data. Then, the long and short-term memory (LSTM) neural network is used to process the time characteristics. Finally, the two are combined to realize the effective processing of the spatio-temporal characteristics of traffic flow data. Experimental results on the real traffic flow dataset prove the feasibility and effectiveness of the proposed method, and can provide a basis for intelligent traffic control and smart city construction. © 2022 The authors and IOS Press.

关键词： Traffic congestion

来源：评论

学校读者我要写书评

暂无评论

超越传统NO_(x)和VOCs控制的臭氧污染防治新思路

引用

science Bulletin 2023年第18期68卷 1989-1992页

作者：陆克定周厚华 James Lee Beth Nelson 张远航 State Environmental Protection Key Laboratory of Atmospheric Ozone Pollution Control State Key Joint Laboratory of Environmental Simulation and Pollution ControlCollege of Environmental Sciences and EngineeringPeking UniversityBeijing 100871China National Centre for Atmospheric Science Department of ChemistryUniversity of YorkYork YO105DDUK

Ambient ozone(O3)was first identified as a key harmful air pollutant in the study of photochemical smog during the 1950s in Los *** studies uncovered that surface O3 was not emitted directly,but formed from the reactions of nitrogen oxides(NO_(x))and volatile organic compounds(VOCs)in *** the relationship between O3 and its two main precursors,NOx and VOCs,is considered to be the cornerstone for enabling successful mitigation of *** the past,a central area for the study and practice of O3 pollution control was Los Angeles and the surrounding *** ambient O3 concentrations in the area were successfully and continuously reduced from 400 ppb(1 ppb=109 m^(3)/m^(3))to ca.120 ppb due to the joint mitigation of VOC and NOx emissions from 1960 to 2010[1].Nevertheless,O3 pollution still represents one of the most pervasive and stubborn environmental problems in megacities worldwide,with many populous centres regularly breaching WHO 8-hour Interim Targets-1 levels(160 lg/m^(3)),not to mention the updated air quality guideline value of 100 lg/m^(3).The difficulty of O3 pollution control is due to a number of factors,including its non-linear relationship with NOx[2],the difficulty of anthropogenic VOC control[3]and the contributions of biogenic VOCs.

关键词： pollution difficulty corners

来源：评论

学校读者我要写书评

暂无评论

Long-time evolution of charged grains in plasma under harmonic external force and after being withdrawn

引用

Chinese Physics B 2022年第2期31卷 460-464页

作者： Miao Guan Zhi-Dong Chen Meng-Die Li Zhong-Mao Liu You-Mei Wang Ming-Yang Yu Department of Physics School of ScienceHangzhou Dianzi UniversityHangzhou 310016China College of Engineering Physics CAMDTShenzhen Key Laboratory of Ultraintense Laser and Advanced Material TechnologyShenzhen Technology UniversityShenzhen 518118China Institute for Fusion Theory and Simulation Zhejiang UniversityHangzhou 310007China Theoretical Physics I Ruhr UniversityBochum D-44780Germany

Evolution of the charged grains in a two-dimensional dusty plasma under a spatially harmonic external force,in particular,their long-time behaviors after the force has been withdrawn,is studied by using molecular dynamics *** an external force and a grain–grain interaction force,initially homogeneously distributed grains can reach a quasistationary state in the form of a disk *** the external force is withdrawn,the disk moves initially with its size and shape nearly unchanged until it rapidly stops moving,and eventually the disk grain rotates like a *** time needed to reach the final state increases with the strength of the initial external force increasing.

关键词： dusty plasma external force long-time behavior molecular dynamics(MD)simulation

来源：评论

学校读者我要写书评

暂无评论

Developing Positively Surface-Charged Carbon Dots as "Semiconductor Electrolyte" for Photo-Electrochemical H2o2 Production Based on Oxygen Reduction

SSRN

引用

SSRN 2023年

作者： Zhang, Yong-Shuai Duan, Hui-Min Wang, Na Wei, Dong-Dong Mo, Liu-Meng Wang, Ning Wang, Xin-Pei Lei, Shuang-Lei Zhang, Yuan Wang, Hong-Yan Key Laboratory for macromolecular Science of Shaanxi Province School of Chemistry and Chemical Engineering Shaanxi Normal University Xi’an710119 China School of Science Key Laboratory of High Performance Scientific Computation Xihua University Chengdu610039 China

Electrolyte modulation for tailoring photoelectrochemical (PEC) activity and selectivity is highly effective but has proved challenging. The major obstacle is lack of a universal strategy that aims for improving the system charge separation based on electrolyte optimization. Herein, we presented an idea as "semiconductor electrolyte", which was developed from the imidazole chloride delivered and positively surface-charged carbon dot (CDs). The CDs aqueous solution can accumulate H2O2 on a CuBi2O4 photocathode via PEC O2 reduction reaction (ORR) at neutral pH without any traditional electrolyte. Combined with experimental measurements and theoretical calculation, the mechanism of system and function of CDs was characterized. As electrolytes, CDs solution enabled to tune the energy band location of CuBi2O4. The electrostatic interaction between positively-charged CDs and negatively-charged photocathode triggered the fast charge and mass migration during PEC proceeding. As semiconductors, CDs can fabricate a dynamic p-n heterojunction with CuBi2O4. These advantages facilitated the photo-induced charge separation, which contributed to the boosted photocurrent and improved stability for PEC system. As ORR occurred interface that incorporated active sites, CDs provided a hydrophilic and adjusted-electronic surface, which fascinated to confine O2 and reduced the kinetic barrier for H2O2 generation. As a result, H2O2 generation was regulated with 0.44 mM and faraday efficiency over 84 % within 10 min, nearly 3 times to the commonly-used Na2SO4 solution. Our work therefore provides a striking and facile strategy to tailor PEC activity and selectivity in a durable and robust pathway. Importantly, it will evoke to develop a universal protocol for regulating PEC performance by electrolyte modulation. © 2023, The Authors. All rights reserved.

关键词： Electrolytes

来源：评论

学校读者我要写书评

暂无评论

Cutting path-dependent machinability of SiCp/Al composite under multi-step ultra-precision diamond cutting

引用

Chinese Journal of Aeronautics 2021年第4期34卷 241-252页

作者： Shijin LU Junjie ZHANG Zengqiang LI Jianguo ZHANG Xiaohui WANG Alexander HARTMAIER Jianfeng XU Yongda YAN Tao SUN Center for Precision Engineering Harbin Institute of TechnologyHarbin 150001China State Key Laboratory of Digital Manufacturing Equipment and Technology School of Mechanical Science and EngineeringHuazhong University of Science and TechnologyWuhan 430074China School of Mechanical Engineering University of JinanJinan 250022China Interdisciplinary Centre for Advanced Materials Simulation Ruhr-University BochumBochum 44780Germany

Particle-tool interactions,which govern the synergetic deformation of SiC particle reinforced Al matrix composites under mechanical machining,strongly depend on the geometry of particle position residing on cutting *** the present work,we investigate the influence of cutting path on the machinability of a SiCp/Al composite in multi-step ultra-precision diamond cutting by combining finite element simulations with experimental observations and *** consistent with experimentally characterized microstructures,the simulated SiCp/Al composite is considered to be composed of randomly distributed polygonally-shaped SiC particles with a volume fraction of 25 vol%.A multi-step cutting strategy with depths of cut ranging from 2 to 10 lm is adopted to achieve an ultimate depth of cut of 10 *** material parameters and extrinsic cutting conditions utilized in finite element simulations of SiCp/Al cutting are consistent with those used in corresponding *** results reveal different particle-tool interactions and failure modes of SiC particles,as well as their correlations with machining force evolution,residual stress distribution and machined surface topography.A detailed comparison between numerical simulation results and experimental data of multi-step diamond cutting of SiCp/Al composite reveals a substantial impact of the number of cutting steps on particle-tool interactions and machined surface *** findings provide guidelines for achieving high surface finish of SiCp/Al composites by ultra-precision diamond cutting.

关键词： Cutting path Diamond cutting Finite element simulation Particle-tool interaction SiCp/Al composites

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：