检索结果-内蒙古大学图书馆

Squeezing More Past Knowledge for Online Class-Incremental Continual Learning

IEEE/CAA Journal of Automatica Sinica 2023年第3期10卷 722-736页

作者： Da Yu Mingyi Zhang Mantian Li Fusheng Zha Junge Zhang Lining Sun Kaiqi Huang the State Key Laboratory of Robotics and System Harbin Institute of Technology(HIT)Harbin 150080China Center for Research on Intelligent System and Engineering Institute of AutomationChinese Academy of Sciences(CASIA)Beijing 100190 School of Artificial Intelligence University of Chinese Academy of SciencesBeijing 100049China Center for Research on Intelligent System and Engineering Institute of AutomationChinese Academy of Sciences(CASIA)Beijing 100190the School of Artificial IntelligenceUniversity of Chinese Academy of SciencesBeijing 100049 the CAS Center for Excellence in Brain Science and Intelligence Technology Shanghai 200031China IEEE

Continual learning(CL)studies the problem of learning to accumulate knowledge over time from a stream of data.A crucial challenge is that neural networks suffer from performance degradation on previously seen data,known as catastrophic forgetting,due to allowing parameter *** this work,we consider a more practical online class-incremental CL setting,where the model learns new samples in an online manner and may continuously experience new ***,prior knowledge is unavailable during training and *** works usually explore sample usages from a single dimension,which ignores a lot of valuable supervisory *** better tackle the setting,we propose a novel replay-based CL method,which leverages multi-level representations produced by the intermediate process of training samples for replay and strengthens supervision to consolidate previous ***,besides the previous raw samples,we store the corresponding logits and features in the ***,to imitate the prediction of the past model,we construct extra constraints by leveraging multi-level information stored in the *** the same number of samples for replay,our method can use more past knowledge to prevent *** conduct extensive evaluations on several popular CL datasets,and experiments show that our method consistently outperforms state-of-the-art methods with various sizes of episodic *** further provide a detailed analysis of these results and demonstrate that our method is more viable in practical scenarios.

关键词： Catastrophic forgetting class-incremental learning continual learning(CL) experience replay

来源：评论

学校读者我要写书评

暂无评论

Numerical Simulation of Streamer Discharge in Transformer Oil Under Various Ionization Energies 19th

Numerical Simulation of Streamer Discharge in Transformer Oi...

引用

19th Annual Conference of China Electrotechnical Society, ACCES 2024

作者： Bao, Yanyan Liang, Shaobin Wang, Feng Zhang, Guangdong Chen, Jianning Zhou, Yuanxiang State Grid Gansu Electric Power Research Institute Lanzhou730070 China State Key Laboratory of Power System Operation and Control Beijing100084 China Department of Electrical Engineering Tsinghua University Beijing100084 China

ISBN: (纸本)9789819613786

Discharge in transformer oil is one of the primary causes of transformer failures. The physical mechanisms of oil discharges are not yet fully understood. Scholars have simulated the discharge process with electrodynamics model, but there is limited research on how the ionization energy, which changes due to various factors such as components and temperature of transformer oil, affect the streamer discharge characteristic. To improve the mechanism of discharge in transformer oil, this paper adopts an electrohydrodynamic model to study the impact of different ionization energies on the streamer discharge characteristics of transformer oil, including initiation voltage, discharge voltage, development speed, and electric field strength. When the ionization energy is between 6.1 eV and 8.1 eV, the initiation voltage and discharge voltage increase linearly with the ionization energy. Within the range of 7.1 eV to 8.1 eV, the time in the streamer accumulation phase increases exponentially with the ionization energy, while the streamer development speed decreases. © Beijing Paike Culture Commu. Co., Ltd. 2025.

关键词： Electrohydrodynamics

来源：评论

学校读者我要写书评

暂无评论

A 1GS/s Highly Linear Voltage-to-Time Converter with Rail-to-Rail Input Range for Time Domain ADCs 16

A 1GS/s Highly Linear Voltage-to-Time Converter with Rail-to...

引用

16th IEEE International Conference on Solid-state and Integrated Circuit Technology, ICSICT 2022

作者： Zhao, Yutong Ye, Fan Ren, Junyan Fudan University State Key Laboratory of Asic and System Department of Microelectronics Shanghai201203 China

ISBN: (数字)9781665469067

ISBN: (纸本)9781665469067

This paper presents a highly linear voltage-to-time converter (VTC) with a bottom-plate sampling and voltage-boost scheme to obtain a rail-to-rail input range. A cascode current mirror and a dynamic threshold crossing detector (TCD) are also added to improve the linearity during the discharging process. Simulated in TSMC 28nm CMOS process, the proposed VTC achieves a 63.6-dB SFDR and an 8.4-bit ENOB for a Nyquist input at 1-GHz sampling frequency while consuming 1.5 mW. The input range is +/-900mV and the output range is +/-662ps. © 2022 IEEE.

关键词： Integrated circuit technology Current mirrors Linearity Detectors CMOS process Time-domain analysis

来源：评论

学校读者我要写书评

暂无评论

OrderMap: A CGRA Mapping Algorithm based on DFG Analysis and Graph Homomorphism 16

OrderMap: A CGRA Mapping Algorithm based on DFG Analysis and...

引用

16th IEEE International Conference on Solid-state and Integrated Circuit Technology, ICSICT 2022

作者： Li, Jiangnan Zheng, Su Wang, Lingli Yin, Wenbo Fudan University State Key Laboratory of Asic and System Department of Microelectronics Shanghai200433 China

ISBN: (数字)9781665469067

ISBN: (纸本)9781665469067

Coarse-grained reconfigurable architecture (CGRA) accelerator is a promising solution in fields such as deep learning and edge computing. One of the difficult problems to design the accelerator is to map the data flow graph (DFG) to the architecture efficiently. The previous mapping algorithms based on graph structure mainly focus on optimization during the placement and routing stage, which may fall into a local minimum and lead to an increase in initiation interval (II). We propose a CGRA mapping algorithm, OrderMap, based on DFG analysis and graph homomorphism. It analyzes the DFG structure information to decide the node mapping order, followed by placement, and routing. Experimental results show that our algorithm is faster than CGRA-ME with smaller IIs. Compared with FastCGRA based on the depth-first order, OrderMap is 22.3 × faster with the same mapping results. © 2022 IEEE.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

Research and Optimization of CSP system Efficiency Based on Effects of Wind

引用

Applied Solar Energy (English translation of Geliotekhnika) 2023年第4期59卷 525-541页

作者： Ali, Kashif Jifeng, Song Department of New Energy North China Electric Power University Beijing China State Key Laboratory of Alternate Electrical Power System with Renewable Energy Sources North China Electric Power University Beijing102206 China

Abstract: Solar thermal power generation has broad development prospects in China’s energy market due to its excellent power quality, continuous power generation, low manufacturing costs, and no pollution to the environment. Based on the theoretical support of computational fluid dynamics, structural strength theory, and Monte Carlo ray tracing method, ray tracing analysis on parabolic trough collector were carried out, to ensure that the concentrating efficiency of collector under the specified wind speed stays within the standard range. Based on the existing parameters, the collector is three-dimensional modeled. The size of the fluid domain was calculated. The calculation model was meshed, and the boundary conditions were set, according to the change of the wind force on the collector under different working angles. The best danger avoidance attitude and the most appropriate maintenance attitude of the collector are obtained, use the data transmission interface between ANSYS and Fluent software to perform a unidirectional fluid-structure coupling analysis on the collector, and pressure-transmit the surface wind pressure of the collector analyzed by Fluent software. Calculate the displacement deformation and equivalent stress distribution of the collector under the effect of wind pressure, Analyze and evaluate its structural strength. The ray tracing software Trace Pro is used to calculate the concentration efficiency under different wind speeds and working angles. Obtain the changing law of the collector efficiency under different wind speeds and different working angles, analyze whether the working efficiency meets the requirements under the two conditions of the design work. © 2023, Allerton Press, Inc.

关键词： Static analysis

来源：评论

学校读者我要写书评

暂无评论

Exploring Self- and Cross-Triplet Correlations for Human-Object Interaction Detection

arXiv

引用

arXiv 2024年

作者： Jiang, Weibo Ren, Weihong Tian, Jiandong Qu, Liangqiong Wang, Zhiyong Liu, Honghai State Key Laboratory of Robotics and System School of Mechanical Engineering and Automation Harbin Institute of Technology Shenzhen China State Key Laboratory of Robotics Shenyang Institute of Automation Chinese Academy of Science China Department of Statistics and Actuarial Science The University of Hong Kong Hong Kong

Human-Object Interaction (HOI) detection plays a vital role in scene understanding, which aims to predict the HOI triplet in the form of . Existing methods mainly extract multi-modal features (e.g., appearance, object semantics, human pose) and then fuse them together to directly predict HOI triplets. However, most of these methods focus on seeking for self-triplet aggregation, but ignore the potential cross-triplet dependencies, resulting in ambiguity of action prediction. In this work, we propose to explore Self- and Cross-Triplet Correlations (SCTC) for HOI detection. Specifically, we regard each triplet proposal as a graph where Human, Object represent nodes and Action indicates edge, to aggregate self-triplet correlation. Also, we try to explore cross-triplet dependencies by jointly considering instance-level, semantic-level, and layout-level relations. Besides, we leverage the CLIP model to assist our SCTC obtain interaction-aware feature by knowledge distillation, which provides useful action clues for HOI detection. Extensive experiments on HICO-DET and V-COCO datasets verify the effectiveness of our proposed SCTC. Copyright © 2024, The Authors. All rights reserved.

关键词： Distillation

来源：评论

学校读者我要写书评

暂无评论

Four decades of full-scale nitrous oxide emission inventory in China

引用

National Science Review 2024年第3期11卷 137-149页

作者： Minqi Liang Zheyan Zhou Peiyang Ren Han Xiao Xu-Ri Zhongmin Hu Shilong Piao Hanqin Tian Qing Tong Feng Zhou Jing Wei Wenping Yuan School of Atmospheric Sciences Guangdong Province Data Center of Terrestrial and Marine Ecosystems Carbon Cycle Sun Yat-sen University State Key Laboratory of Tibetan Plateau Earth System Resources and Environment Institute of Tibetan Plateau Research Chinese Academy of Sciences Key Laboratory of Agro-Forestry Environmental Processes and Ecological Regulation of Hainan Province Hainan University Sino-French Institute for Earth System Science College of Urban and Environmental SciencesPeking University Schiller Institute for Integrated Science and Society Department of Earth and Environmental Sciences Boston College Institute of Energy Environment and Economy Tsinghua University Southern Marine Science and Engineering Guangdong Laboratory(Zhuhai)

China is among the top nitrous oxide（N2O）-emitting countries, but existing national inventories do not provide full-scale emissions including both natural and anthropogenic sources. We conducted a four-decade（1980–2020） of comprehensive quantification of Chinese N2O inventory using empirical emission factor method for anthropogenic sources and two up-to-date process-based models for natural sources. Total N2O emissions peaked at 2287.4（1774.8–2799.9） Gg N2O yr-1in 2018, and agriculture-developed regions, like the East, Northeast, and Central, were the top N2O-emitting regions. Agricultural N2O emissions have started to decrease after 2016 due to the decline of nitrogen fertilization applications, while, industrial and energetic sources have been dramatically increasing after 2005. N2O emissions from agriculture, industry,energy, and waste represented 49.3%, 26.4%, 17.5%, and 6.7% of the anthropogenic emissions in 2020,respectively, which revealed that it is imperative to prioritize N2O emission mitigation in agriculture,industry, and energy. Natural N2O sources, dominated by forests, have been steadily growing from 317.3（290.3–344.1） Gg N2O yr-1in 1980 to 376.2（335.5–407.2） Gg N2O yr-1in 2020. Our study produces a Full-scale Annual N2O dataset in China（FAN2020）, providing emergent counting to refine the current national N2O inventories.

关键词： nitrous oxide four decades anthropogenic sources natural sources

来源：评论

学校读者我要写书评

暂无评论

A Novel Gait Pattern Recognition Method Based on LSTM-CNN for Lower Limb Exoskeleton

引用

Journal of Bionic Engineering 2021年第5期18卷 1059-1072页

作者： Chao-feng Chen Zhi-jiang Du Long He Yong-jun Shi Jia-qi Wang Wei Dong State Key Laboratory of Robotics and System Harbin Institute of Technology(HIT)Harbin 150001China China South Industries Group Corporation Weapon Equipment Research InstituteBeijing 102202China

This paper describes a novel gait pattern recognition method based on Long Short-Term Memory(LSTM)and Convolutional Neural Network(CNN)for lower limb *** Inertial Measurement Unit(IMU)installed on the exoskeleton to collect motion information,which is used for LSTM-CNN *** article considers five common gait patterns,including walking,going up stairs,going down stairs,sitting down,and standing *** the LSTM-CNN model,the LSTM layer is used to process temporal sequences and the CNN layer is used to extract *** optimize the deep neural network structure proposed in this paper,some hyperparameter selection experiments were carried *** addition,to verify the superiority of the proposed recognition method,the method is compared with several common methods such as LSTM,CNN and *** results show that the average recognition accuracy can reach 97.78%,which has a good recognition eff ***,according to the experimental results of gait pattern switching,the proposed method can identify the switching gait pattern in time,which shows that it has good real-time performance.

关键词： Lower limb exoskeleton Gait pattern recognition LSTM-CNN Recognition accuracy Real-time performance

来源：评论

学校读者我要写书评

暂无评论

NNPP: A Learning-Based Heuristic Model for Accelerating Optimal Path Planning on Uneven Terrain

arXiv

引用

arXiv 2023年

作者： Ji, Yiming Liu, Yang Xie, Guanghu Ma, Boyu Xie, Zongwu Cao, Baoshi State Key Laboratory of Robotics and System Harbin Institute of Technology Heilongjiang Harbin150001 China

Intelligent autonomous path planning is essential for enhancing the exploration efficiency of mobile robots operating in uneven terrains like planetary surfaces and off-road environments. In this paper, we propose the NNPP model for computing the heuristic region, enabling foundation algorithms like A* to find the optimal path solely within this reduced search space, effectively decreasing the search time. The NNPP model learns semantic information about start and goal locations, as well as map representations, from numerous pre-annotated optimal path demonstrations, and produces a probabilistic distribution over each pixel representing the likelihood of it belonging to an optimal path on the map. More specifically, the paper computes the traversal cost for each grid cell from the slope, roughness and elevation difference obtained from the digital elevation model. Subsequently, the start and goal locations are encoded using a Gaussian distribution and different location encoding parameters are analyzed for their effect on model performance. After training, the NNPP model is able to accelerate path planning on novel maps. Experiments demonstrate that the heuristic region generated by the NNPP model achieves a 3×speedup for optimal path planning under identical hardware conditions. Moreover, the NNPP model’s advantage becomes more pronounced as the size of the map increases. Copyright © 2023, The Authors. All rights reserved.

关键词： Motion planning

来源：评论

学校读者我要写书评

暂无评论

A 12-Bit 250-MS/s Voltage-Time Domain Hybrid ADC 16

A 12-Bit 250-MS/s Voltage-Time Domain Hybrid ADC

引用

16th IEEE International Conference on Solid-state and Integrated Circuit Technology, ICSICT 2022

作者： Zhao, Yutong Ye, Fan Ren, Junyan Fudan University State Key Laboratory of Asic and System Department of Microelectronics Shanghai201203 China

ISBN: (数字)9781665469067

ISBN: (纸本)9781665469067

This article presents a 12-bit 250-MS/s voltage-time domain hybrid analog-to-digital converter (ADC). In the proposed ADC, the first stage is a binary-weighted SAR ADC, and the second stage is a coarse-fine time-to-digital converter (TDC), connected with a highly linear dynamic voltage-to-time converter (VTC). Using the combination of voltage and time domain quantization, this hybrid architecture is power efficient and can benefit from technology scaling. The time domain folding of the ring oscillator leads to an inherently linear operation. The prototype ADC is simulated in TSMC 28nm CMOS process with a 0.9V power supply, achieving a 55.62-dB SNDR and a 67.16-dB SFDR at Nyquist input. The power consumption is 1.36 mW, obtaining a Walden figure of merit (FoM) of 11.07 fJ/conv.- step. © 2022 IEEE.

关键词： Frequency converters

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：