检索结果-内蒙古大学图书馆

IEEE Transactions on Circuits and systems for Artificial Intelligence 2024年第1期1卷 37-52页

作者： Viviane Potocnik Luca Colagrande Tim Fischer Luca Bertaccini Daniele Jahier Pagliari Alessio Burrello Luca Benini Integrated Systems Laboratory (IIS) of ETH Zürich ETZ Züurich Switzerland Department of Control and Computer Engineering Politecnico di Torino Turin Italy Department of Electrical Electronic and Information Engineering University of Bologna Bologna Italy

Transformer-based foundation models have become crucial for various domains, most notably natural language processing (NLP) or computer vision (CV). These models are predominantly deployed on high-performance GPUs or hardwired accelerators with highly customized, proprietary instruction sets. Until now, limited attention has been given to RISC-V-based general-purpose platforms. In our work, we present the first inference results of transformer models on an open-source many-tiny-core RISC-V platform implementing distributed Softmax primitives and leveraging ISA extensions for SIMD floating-point operand streaming and instruction repetition, as well as specialized DMA engines to minimize costly main memory accesses and to tolerate their latency. We focus on two foundational transformer topologies, encoder-only and decoder-only models. For encoder-only models, we demonstrate a speedup of up to 12.8 $\times$ between the most optimized implementation and the baseline version. We reach over 79% FPU utilization and 294 GFLOPS/W, outperforming State-of-the-Art (SoA) accelerators by more than 2 $\times$ utilizing the HW platform while achieving comparable throughput per computational unit. For decoder-only topologies, we achieve 16.1 $\times$ speedup in the Non-Autoregressive (NAR) mode and up to 35.6 $\times$ speedup in the Autoregressive (AR) mode compared to the baseline implementation. Compared to the best SoA dedicated accelerator, we achieve 2.04 $\times$ higher FPU utilization.

关键词： Frequency modulation Transformers computer architecture Computational modeling Kernel Artificial intelligence Hardware

来源：评论

学校读者我要写书评

暂无评论

Exploring the use of deep learning in task-flexible ILC*

Exploring the use of deep learning in task-flexible ILC*

引用

American control Conference (ACC)

作者： Anantha Sai Hariharan Vinjarapu Yorick Broens Hans Butler Roland Tóth Department of Electrical Engineering Eindhoven University of Technology Eindhoven The Netherlands ASML Veldhoven The Netherlands Systems and Control Laboratory Institute for Computer Science and Control Hungary

Growing demands in today’s industry results in increasingly stringent performance and throughput specifications. For accurate positioning of high-precision motion systems, feedforward control plays a crucial role. Nonetheless, conventional model-based feedforward approaches are no longer sufficient to satisfy the challenging performance requirements. An attractive method for systems with repetitive motion tasks is iterative learning control (ILC) due to its superior performance. However, for systems with non-repetitive motion tasks, ILC is generally not applicable, despite of some recent promising advances. In this paper, we aim to explore the use of deep learning to address the task flexibility constraint of ILC. For this purpose, a novel Task Analogy based Imitation Learning (TAIL)-ILC approach is developed. To benchmark the performance of the proposed approach, a simulation study is presented which compares the TAIL-ILC to classical model-based feedforward strategies and existing learning-based approaches, such as neural network based feedforward learning.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Binary Stochastic Flip Optimization for Training Binary Neural Networks

Binary Stochastic Flip Optimization for Training Binary Neur...

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Tatsukichi Shibuya Nakamasa Inoue Rei Kawakami Ikuro Sato Department of Computer Science Institute of Science Tokyo Tokyo Japan Department of Systems and Control Engineering Institute of Science Tokyo Tokyo Japan Denso IT Laboratory Tokyo Japan

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

For deploying deep neural networks on edge devices with limited resources, binary neural networks (BNNs) have attracted significant attention, due to their computational and memory efficiency. However, once a neural network is binarized, finetuning it on edge devices becomes challenging because most conventional training algorithms for BNNs are designed for use on centralized servers and require storing real-valued parameters during training. To address this limitation, this paper introduces binary stochastic flip optimization (BinSFO), a novel training algorithm for BNNs. BinSFO employs a parameter update rule based on Boolean operations, eliminating the need to store real-valued parameters and thereby reducing memory requirements and computational overhead. In experiments, we demonstrated the effectiveness and memory efficiency of BinSFO in fine-tuning scenarios on six image classification datasets. BinSFO performed comparably to conventional training algorithms with a 70.7% smaller memory requirement. Code is released at https://***/TatsukichiShibuya/ICASSP2025_BinSFO

关键词： Training Speech coding Memory management Signal processing algorithms Signal processing Classification algorithms Servers Speech processing Optimization Image classification

来源：评论

学校读者我要写书评

暂无评论

Exploring the use of deep learning in task-flexible ILC

arXiv

引用

arXiv 2023年

作者： Vinjarapu, Anantha Sai Hariharan Broens, Yorick Butler, Hans Tóth, Roland The Department of Electrical Engineering Eindhoven University of Technology Eindhoven Netherlands The Systems and Control Laboratory Institute for Computer Science and Control Hungary Asml Veldhoven Netherlands

Growing demands in today's industry results in increasingly stringent performance and throughput specifications. For accurate positioning of high-precision motion systems, feedforward control plays a crucial role. Nonetheless, conventional model-based feedforward approaches are no longer sufficient to satisfy the challenging performance requirements. An attractive method for systems with repetitive motion tasks is iterative learning control (ILC) due to its superior performance. However, for systems with non-repetitive motion tasks, ILC is generally not applicable, despite of some recent promising advances. In this paper, we aim to explore the use of deep learning to address the task flexibility constraint of ILC. For this purpose, a novel Task Analogy based Imitation Learning (TAIL)-ILC approach is developed. To benchmark the performance of the proposed approach, a simulation study is presented which compares the TAIL-ILC to classical model-based feedforward strategies and existing learning-based approaches, such as neural network based feedforward *** Codes 37M15 © 2023, CC BY.

关键词： Feedforward control

来源：评论

学校读者我要写书评

暂无评论

A SINS Error Correction Approach Based on Dual-Threshold ZV Detection and Cubature Kalman Filter

A SINS Error Correction Approach Based on Dual-Threshold ZV ...

引用

2023 IEEE International Conference on systems, Man, and Cybernetics, SMC 2023

作者： Xu, Ruijie Chen, Shichao Sun, Wenqiao Lv, Yisheng Luo, Jialiang Tang, Ying Institute of Automation Chinese Academy of Sciences College of Information Science & Technology Beijing University of Chemical Technology The State Key Laboratory for Management and Control of Complex System State Key Laboratory of Multimodal Artificial Intelligence Systems Beijing China Institute of Automation Chinese Academy of Sciences The Center of National Railway Intelligent Transportation System Engineering and Technology China Academy of Railway Sciences Corporation Limited The State Key Laboratory for Management and Control of Complex System State Key Laboratory of Multimodal Artificial Intelligence Systems Beijing China Transportation and Economics Research Institute The Center of National Railway Intelligent Transportation System Engineering and Technology China Academy of Railway Sciences Corporation Limited Beijing China Institute of Automation Chinese Academy of Sciences The State Key Laboratory for Management and Control of Complex System State Key Laboratory of Multimodal Artificial Intelligence Systems Beijing China Institute of Automation Chinese Academy of Sciences China University of Geosciences Beijing School of Information Engineering The State Key Laboratory for Management and Control of Complex System State Key Laboratory of Multimodal Artificial Intelligence Systems Beijing China Rowan University Department of Electrical and Computer Engineering Glassboro United States

ISBN: (纸本)9798350337020

Global Navigation Satellite systems (GNSS) can provide real-time positioning information for outdoor users, but cannot for indoor scenarios or heavily occluded outdoor scenarios. Strap-down Inertial Navigation System (SINS) are widely used to locate people in complex interior or heavily occluded outdoor scenarios due to its light weight and low power consumption. However, IMU of SINS are noisy, and the sampling data error is large, which is a divergence of the error with time. Therefore, it will generate a positioning accumulation error, which affects the final positioning accuracy. The problem of cumulative IMU errors is usually dealt with by Zero-Velocity Update (ZUPT). The zero-velocity detection part of basic ZUPT method usually uses a single threshold to determine the gait of pedestrian, which often has the problem of gait misjudgment and omission. To address these problems, this paper proposes a composite conditional detection method to solve the problem of misjudgment in the zero-velocity interval. In addition, we redesign the zero-velocity update algorithm and uses the Cubature Kalman filter (CKF) for pedestrian positioning error correction. The experimental results demonstrate that the proposed ZUPT method based on dual-threshold detection can better detect the interval between pedestrian motion and stationery than ones with single threshold. The zero-velocity update algorithm based on CKF has higher performance than conventional EKF and UKF methods, which constrains the cumulative error of SINS to about 0.2% of the whole walking distance. © 2023 IEEE.

关键词： Cubature Kalman Filter Inertial Pedestrian Navigation Zero-Velocity Update

来源：评论

学校读者我要写书评

暂无评论

Improving Motor Imagery Classification in BCI systems Using EMD and Multi-Layer CNNs

Improving Motor Imagery Classification in BCI Systems Using ...

引用

International eConference on computer and Knowledge engineering (ICCKE)

作者： Reza Arghand Ali Chaibakhsh Moein Radman Intelligent Systems and Advanced Control Lab Faculty of Mechanical Engineering University of Guilan Rasht Guilan Iran Brain-Computer Interfacing and Neural Engineering Laboratory University of Essex School of Computer Science and Electronic Engineering Colchester United Kingdom

ISBN: (数字)9798331511272

ISBN: (纸本)9798331511289

Brain-computer Interface (BCI) systems are relatively new technologies that could play a significant role in aiding the recovery of impaired activities resulting from neuromuscular disabilities in affected individuals. Accurate recognition and classification of motor imagery in BCI systems present a challenge, leading to extensive research in recent years aimed at improving the accuracy of these systems. In this study, a combination of the Empirical Mode Decomposition (EMD) method and a multi-layer Convolutional Neural Network (CNN) is employed. Initially, the signal is decomposed into Intrinsic Mode Functions (IMFs) using EMD, and all IMFs across all trials are analyzed to select the best ones, which are then fed into the CNN as inputs. Additionally, the study incorporates the fusion of three CNN networks, each corresponding to a different IMF. The features extracted from these networks are combined and used to train an SVM classifier. The proposed method achieved an accuracy of 86.1% on the BCI-2a 2008 dataset, outperforming other state-of-the-art approaches.

关键词： Support vector machines Knowledge engineering Accuracy Image recognition Empirical mode decomposition Neuromuscular Motors Feature extraction Brain-computer interfaces Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Optimal Sensor Placement for Full-Set TDOA Localization Accounting for Sensor Location Errors

引用

IEEE Transactions on Aerospace and Electronic systems 2025年

作者： Zhang, Chengjie Han, Xinyang Kong, He Ho, K.C. Southern University of Science and Technology Shenzhen Key Laboratory of Control Theory and Intelligent Systems Guangdong Provincial Key Laboratory of Fully Actuated System Control Theory and Technology Shenzhen518055 China University of Missouri Electrical Engineering and Computer Science Department ColumbiaMO65211 United States

The positioning accuracy of a signal source is influenced by where the sensors are deployed. Studies in the literature for the optimal sensor placement (OSP) of localization often ignore the presence of sensor position errors. In practice, however, it is impossible to place the sensors precisely at the desired locations. This work designs OSP by accounting for the presence of Gaussian sensor position errors that are independent and identically distributed (IID), with the full set of Time Difference of Arrival measurements corrupted by Gaussian IID noise. The criterion is the A-optimality equal to the trace of the source position Cramer-Rao Bound (CRB). We show that the presence of sensor position errors introduces a multiplication factor to the same criterion for the case without sensor position errors. The latter factor is dependent on the ratio of the sensor position error power to the range difference noise power but independent of the sensor positions, implying that the A-optimality OSP in the presence of sensor position errors is the same as in the absence. To our knowledge, this is the first time such results have been derived. Simulation studies validate the derived OSP, show the minimal OSP performance gap with sensor positioning error and without, and illustrate the better source localization accuracy by OSP. © 1965-2011 IEEE.

关键词： Cramer-Rao bounds

来源：评论

学校读者我要写书评

暂无评论

Estimating the Sensitivity of Microoptoelectromechanical Micro-g Accelerometer 23

Estimating the Sensitivity of Microoptoelectromechanical Mic...

引用

23rd IEEE International Conference of Young Professionals in Electron Devices and Materials, EDM 2022

作者： Barbin, Evgenij S. Nesterenko, Tamara G. Koleda, Aleksej N. Shesterikov, Evgeniy V. Kulinich, Ivan V. Sheyerman, Fedor I. The Tomsk State University of Control Systems and Radioelectronics Laboratory of Radiophotonics Tomsk Russia Institute of Optics and Atmosphere Siberian Branch of the Russian Academy of Sciences Laboratory of Intelligent Computer Systems Tomsk Russia National Research Tomsk Polytechnic University Division for Electronic Engineering Tomsk Russia

ISBN: (纸本)9781665498043

The article studies a MOEMS-accelerometer with an optical measuring element based on an optical tunneling effect intended for measuring micro-g accelerations. The work estimates the sensitivity threshold of the accelerometer based on the Noise Equivalent Acceleration, which is stipulated by thermal Brownian motion of molecules, noises of a laser and photodiode. Shot noise has the largest effect on the MOEMS-accelerometer noise threshold and depends on its mechanical properties. The general sensitivity of the MOEMS-accelerometer equals 0.938, 3, or 6.255 A/g, depending on a chosen operational point. The sensitivity threshold in terms of photodiode dark current equals 107 × 10-9 g, 33.3 × 10-9 g, 16 × 10-9 g. The dynamic range for the photodiode equals ± 3.2 × 10-3 g, while for the measuring transducer it is ±0.32g, ±0.1, or ±0.05g. To decrease the noise down to 0.1 µ g, the inertial mass should amount at least to 65 × 10-7 kg, and the average laser noise power should be below 1 %. © 2022 IEEE.

关键词： Accelerometers

来源：评论

学校读者我要写书评

暂无评论

Consensus control of Leader-Following Multi-Agent systems in Directed Topology With Heterogeneous Disturbances

引用

IEEE/CAA Journal of Automatica Sinica 2021年第2期8卷 423-431页

作者： Qinglai Wei Xin Wang Xiangnan Zhong Naiqi Wu the State Key Laboratory for Management and Control of Complex Systems Institute of AutomationChinese Academy of SciencesBeijing 100190 the School of Artificial Intelligence University of Chinese Academy of SciencesBeijing 100049and also with Qingdao Academy of Intelligent IndustriesQingdao 266109China the Department of Computer and Electrical Engineering and Computer Science Florida Atlantic UniversityBoca Raton FL33431 USA the Institute of Systems Engineering and Collaborative Laboratory for Intelligent Science and SystemsMacao University of Science and TechnologyMacao 999078China

This paper investigates the consensus problem for linear multi-agent systems with the heterogeneous disturbances generated by the Brown *** main contribution is that a control scheme is designed to achieve the dynamic consensus for the multi-agent systems in directed topology interfered by stochastic *** traditional ways,the coupling weights depending on the communication structure are static.A new distributed controller is designed based on Riccati inequalities,while updating the coupling weights associated with the gain matrix by state errors between adjacent *** introducing time-varying coupling weights into this novel control law,the state errors between leader and followers asymptotically converge to the minimum value utilizing the local *** the Lyapunov directed method and It?formula,the stability of the closed-loop system with the proposed control law is *** simulation results conducted by the new and traditional schemes are presented to demonstrate the effectiveness and advantage of the developed control method.

关键词： Consensus control directed topology external disturbance multi-agent(MA)systems

来源：评论

学校读者我要写书评

暂无评论

Geometric Parameterized Macromodel of Uniaxial Capacitance Accelerometer Based on Modal Superposition 16

Geometric Parameterized Macromodel of Uniaxial Capacitance A...

引用

16th IEEE International Scientific and Technical Conference Actual Problems of Electronic Instrument engineering, APEIE 2023

作者： Gridchin, Aleksandr V. Koleda, Aleksei N. Barbin, Evgenii S. Nesterenko, Tamara G. Kolchuzhin, Vladimir A. Department of Semiconductor Devices and Microelectronics Novosibirsk State Technical University Novosibirsk Russia Department of Electronic Engineering National Research Tomsk Polytechnic University Tomsk Russia Laboratory of Intelligent Computer Systems Tomsk State University of Control Systems and Radioelectronics Tomsk Russia Advanced Cellular Group Qorvo Munich GmbH Munich Germany

ISBN: (纸本)9798350330885

The paper presents a piecewise geometric parameterization method and a parametric analysis of uniaxial capacitance accelerometer using a reduced model formulated on the modal superposition method. The extraction of parameters for the macromodel such as modal frequencies, modal masses, modal stiffnesses, strain energy, and capacitances are described. The reduced model consists of the two modes: torsion mode, and transversal mode, three electrodes in the electrostatic domain, and eight master nodes in the geometrical domain located at the edges of the plate. The generation of the discrete macromodels is executed using commercial software. The thickness of the plate has been selected as a design parameter. Linear piecewise interpolation is used for macromodel parameterization which have been done inside VHDL-AMS code. The accelerometer's dynamic response simulated using of parametrized macromodel considering the mechanical contact phenomena and the nonlinear electrostatic interaction are presented and discussed. The developed approach can be applied for fast system level simulations in view of the variation of the geometric parameters. © 2023 IEEE.

关键词： Capacitance

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：