检索结果-南通市图书馆

Channel pruning on frequency response

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2025年第1期68卷 159-170页

作者： Hang LIN Yifan PENG Lin BIE Chenggang YAN Xibin ZHAO Yue GAO School of Software Tsinghua University School of Automation Hangzhou Dianzi University

Network pruning has a significant role in reducing network parameters and accelerating the inference time of the network. Some existing methods prune the network based on the frequency of the data, and finally obtain a sub-network with high accuracy. However, according to our experimental analysis, different frequencies of information in the data contribute differently to the accuracy of the model, and using this information directly for pruning without making a selection will lead to incorrect results. We believe that pruning should retain the convolutional kernels in the network that process important information, while those kernels that process unimportant information should be removed. In this paper, we first investigate the meaning of each frequency band information in the spectrum and their contribution to the prediction accuracy of the network,and according to these results, we propose a new pruning method based on frequency response(PFR). Our PFR finds and removes the convolutional kernels in the network that specialize in processing unimportant information, resulting in a compact neural network model. PFR obtains significant experimental results on different datasets, for example, a 56.0% raduction of float points operations(FLOPs) on Res Net-50 and only 0.3.% of Top-1 accuracy degradation on the Image Net dataset.

关键词： deep learning model compression filter pruning channel pruning frequency response

同方期刊数据库

Software-defined process-near-memory architecture using 3. hybrid bonding integration

在线全文

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2025年第1期68卷 366-378页

作者： Anlin XU Chenchen DENG Jianfeng ZHU Yao WANG Shaojun WEI Leibo LIU School of Integrated Circuits Tsinghua University Beijing Institution of Tracking and Communication Technology Beijing National Research Center for Information Science and Technology Tsinghua University

With the unprecedented explosive growing amount of global data, the development of computing chips, which encounter bottlenecks such as power wall and memory wall, cannot satisfy the demanding requirement. This work proposes a software-defined process-near-memory(SDPNM) computing architecture implemented using 3. hybrid bonding *** software-defined chip architecture, featuring spatial computations and dynamic reconfiguration, innovates in a top-down manner to achieve high energy efficiency while maintaining flexibility after fabrication. The process-near-memory integration further advances the SDNPM chip in a bottom-up way to reduce the energy consumption of data movement while improving the bandwidth. Utilizing a relatively mature fabrication and bonding process can result in feasible solutions for both data-intensive and compute-intensive applications including digital signal processing and artificial intelligence. The logic die is fabricated in the SMIC 40 nm process and the DRAM die is fabricated in the PSMC 25 nm process. The hybrid bonding is implemented by XMC. The experimental results show that the energy efficiency of the proposed SDPNM chip is 3..1× better than the state-of-the-art FPGA ranging from 8.2× to 104.1×.

关键词： software-defined chips process near memory energy efficiency dynamic reconfiguration domain specific

同方期刊数据库

Identifiability and sloppiness of structured systems with a matrix fraction description using finite frequency responses

在线全文

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2025年

作者： Yunxiang MA Tong ZHOU Department of Automation Tsinghua University Department of Automation and BNRist Tsinghua University

This paper investigates the identifiability of structured systems with a matrix fraction description and attacks practical difficulties in estimating parameters under finite frequency responses. The numerator and denominator polynomial matrices depend affinely on system parameters, which can represent a broad range of systems, especially circuits and mechanical systems. Except for assuming that the denominator matrix is invertible, no further assumptions are adopted. A criterion is provided for validating this assumption utilizing graph theories and matroids. A sufficient and necessary criterion for global identifiability based on the rank of a numerical matrix is given and can be recursively verified with each frequency point,which is computationally attractive for large-scale systems. For an identifiable system, an ellipsoid approximation is obtained for the set of parameters whose corresponding system frequency responses deviate within a specific range from those of the system at a specific parameter value. Using this approximation, explicit expressions for absolute and relative sloppiness have been derived, which can be used to quantify the difficulty of identifying system parameters. Comparisons with the well-known Fisher information matrix have also been performed through a mechanical system, revealing significant differences in quantifying parameter estimation hardness.

关键词：

CO2-mediated bicarbonate conversion to concentrated formate in a CEM-based electrolyzer

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Journal of Energy Chemistry 2025年第1期100卷 605-611页

作者： Haocheng Xiong Donghuan Wu Haonan Li Andrew Li Qikun Hu Siyao Song Bingjun Xu Qi Lu State Key Laboratory of Chemical Engineering Department of Chemical Engineering Tsinghua University College of Chemistry and Molecular Engineering Peking University Ordos Laboratory

Renewable energy-driven bicarbonate conversion to valuable chemicals presents an attractive strategy for mitigating CO2emissions,as bicarbonate can be efficiently generated from the capture of atmospheric CO2using alkaline solutions with reactive *** this work,we present a CO2-mediated bicarbonate conversion to pure formate using a cation exchange membrane-based electrolyzer with a 25 cm2electrode *** electrolysis achieved selectivities exceeding 75%for formate at a total current of 2.5 A,achieving formate concentrations up to 1.2 M and yields as high as 95%over extended *** techno-economic assessment confirmed the economic viability of the process,highlighting the potential for bicarbonate electrolysis as a sustainable method for producing valuable chemicals.

关键词： Electrocatalysis Techno-economic assessments Electrochemistry Flow electrolyzer

Targets capture by distributed active swarms via bio-inspired reinforcement learning

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Physics,Mechanics & Astronomy) 2025年第1期Mechanics & Astronomy) .卷 210-221页

作者： Kun Xu Yue Li Jun Sun Shuyuan Du Xinpeng Di Yuguang Yang Bo Li Institute of Biomechanics and Medical Engineering Applied Mechanics Laboratory Department of Engineering MechanicsTsinghua University Institute of Nuclear and New Energy Technology Tsinghua University Shanghai Aerospace Control Technology Institute Shanghai Key Laboratory of Aerospace Intelligent Control Technology College of Automation Engineering Nanjing University of Aeronautics and Astronautics

Natural swarms arranged from cells to herds are usually decentralized but display intriguing collective intelligence in coordinating individuals across large scales to efficiently achieve their common goals. Learning from nature may provide new strategies for controlling collective dynamics of synthetic swarms to accomplish specific functions. Here, we present a bioinspired computational framework that steers distributed active swarms to collectively capture and merge targets via reinforcement learning. We exploit collective milling structures of natural herds to cage the targets, and adopt a switching control policy inspired by sperms' chiral dynamics to optimize the trajectories of individuals, through which the active swarms can selforganize to enclose single or multiple distant targets in a dynamical, adaptive and scalable manner. There exists a critical swarm size, beyond which the excessive competition between agents would generate large mechanical forces, leading to capture instability but enabling the transition from short-distance to long-distance merging capture of multiple targets. This work provides physical insights into distributed active swarms and could offer a multilevel, decentralized strategy toward controlling swarm robotics in wide applications such as bio-medical devices, machine immunity, and target clearance.

关键词： swarm intelligence robotics multi-tasks artificial intelligence bionics

Emerging trends in the selective production of value-added aromatics

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China Chemistry 2025年第1期68卷 15-20页

作者： Yuchao Wu Houliang Dai Weisheng Yang Yaxing Li Lichen Liu Petrochemical Research Institute Company Limited China National Petroleum Corporation Engineering Research Center of Advanced Rare Earth Materials Department of Chemistry Tsinghua University

1 Introduction Aromatics (mainly benzene, toluene, and xylenes, BTX) are essential intermediates in the chemical industry. They are employed in the synthesis of a broad spectrum of chemicals and materials, e.g., various polymers, plastics, resins, solvents, dyes, pharmaceuticals [1]. The manufacturing pathway of BTX mainly includes petroleum reforming of naphtha, coal-to-aromatics, methanol-to-aromatics [2].

关键词：

Revealing and predicting the long-term biomechanical response of orthokeratology by developing a patient-specific computational model

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Physics,Mechanics & Astronomy) 2025年第1期Mechanics & Astronomy) .卷 156-168页

作者： Yifeng Li Zhuoran Yang Ziming Yan Huibin Shi Zhanli Liu Kaijie Wang Department of Engineering Mechanics School of Aerospace Engineering Tsinghua University Beijing Tongren Eye Center Beijing Tongren Hospital Capital Medical University Beijing Ophthalmology & Visual Sciences Key Lab

Orthokeratology(OK) is widely used for effective myopia correction and control. However, the incomplete understanding of its biomechanical mechanisms makes OK lens fitting rely heavily on clinician judgment, complicating accurate predictions of treatment outcomes. In this paper, we performed clinical experiments and numerical analysis to study corneal deformation modes and long-term changes in central corneal thickness. Clinical experiments were conducted on 194 Chinese myopic patients under OK treatment for 3.months. Based on the experimental data, a patient-specific computational biomechanical model for OK was established and validated. Specifically, the anisotropic mechanical properties of the cornea were incorporated into the model to describe the significant difference between its shear modulus(29.5 kPa) and tensile modulus(768.4 kPa). Additionally, a viscohyperelastic material model with a prolonged corneal relaxation time of 5.6 h was developed to capture the long-term deformation response. The results show that corneal thickness reduction in OK is primarily due to out-of-plane shear deformation,influenced by the cornea's low shear resistance. Modeling the extended corneal relaxation time is crucial for predicting long-term biomechanical responses. The computational model effectively captures long-term changes in central corneal thickness, potentially improving OK lens fitting accuracy.

关键词： orthokeratology computational biomechanical model long-term viscoelasticity myopia correction

A Bayesian expectation maximization algorithm for state estimation of intelligent vehicles considering data loss and noise uncertainty

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China Technological Sciences 2025年

作者： Yan WANG Feng TIAN Jianqiang WANG Keqiang LI Department of Industrial and Systems Engineering The Hong Kong Polytechnic University School of Vehicle and Mobility Tsinghua University

Sideslip angle, yaw rate, and vehicle velocity are essential for intelligent vehicle control. Since these vehicle states are not measured directly, some Kalman-based approaches have been developed to estimate these states using in-vehicle sensors. However, the existing studies seldom account for the influence of sensor data loss on estimation accuracy. In addition,the process and measurement noise change during the estimation process because of the various driving conditions. To address these problems, an expectation-maximization robust extended Kalman filter (EMREKF) is proposed. Firstly, a robust extended Kalman filter (REKF) is developed to deal with the impact of missing measurements. Then, an improved expectation maximization (EM) algorithm that considers data loss is presented to update the noise parameter of the REKF dynamically. Finally,the improved EM is fused with the REKF to form the EMREKF to estimate vehicle state. The experimental results demonstrate that the EMREKF outperforms EKF, REKF, and maximum correntropy criterion EKF for various degrees of data loss and the proposed algorithm has a strong adaptive ability to different driving conditions.

关键词：

Recent Advances in Artificial Sensory Neurons:Biological Fundamentals,Devices,Applications,and Challenges

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Nano-Micro Letters 2025年第3期17卷 168-216页

作者： Shuai Zhong Lirou Su Mingkun Xu Desmond Loke Bin Yu Yishu Zhang Rong Zhao Guangdong Institute of Intelligence Science and Technology HengqinZhuhai 519031People’s Republic of China College of Integrated Circuits Zhejiang UniversityHangzhou 3112000People’s Republic of China ZJU-Hangzhou Global Scientific and Technological Innovation Center Hangzhou 310027People’s Republic of China Department of Science Mathematics and TechnologySingapore University of Technology and DesignSingapore 487372Singapore Department of Precision Instruments Tsinghua UniversityBeijing 100084People’s Republic of China Center for Brain‑Inspired Computing Research Tsinghua UniversityBeijing 100084People’s Republic of China IDG/McGovern Institute for Brain Research Tsinghua UniversityBeijing 100084People’s Republic of China

Spike-based neural networks,which use spikes or action potentialsto represent information,have gained a lot of attention because of their high energyefficiency and low power *** fully leverage its advantages,convertingthe external analog signals to spikes is an essential *** including analog-to-digital converters or ring oscillators,and sensorssuffer from high power and area *** efforts are devoted to constructingartificial sensory neurons based on emerging devices inspired by the biologicalsensory *** can simultaneously perform sensing and spike conversion,overcoming the deficiencies of traditional sensory *** review summarizesand benchmarks the recent progress of artificial sensory *** starts with thepresentation of various mechanisms of biological signal transduction,followed bythe systematic introduction of the emerging devices employed for artificial ***,the implementations with different perceptual capabilitiesare briefly outlined and the key metrics and potential applications are also ***,we highlight the challenges and perspectives for the future development of artificial sensory neurons.

关键词： Artificial intelligence Emerging devices Artificial sensory neurons Spiking neural networks Neuromorphic sensing

维普期刊数据库