检索结果-内蒙古大学图书馆

AQSS: Accelerator of quantization neural networks with stochastic approach 6

AQSS: Accelerator of quantization neural networks with stoch...

6th International Symposium on Computing and Networking (CANDAR) - Across Practical Development and Theoretical Research

作者： Ueki, Takeo Iwai, Keisuke Matsubara, Takashi Kurokawa, Takakazu Natl Def Acad Japan Dept Comp Sci Hashirimizu 1-10-20 Yokosuka Kanagawa 2390811 Japan

ISBN: (纸本)9781538691847

In recent years, Deep Neural Network (DNN) s have become widely spread. Several high-throughput hardware implementations for DNNs have been proposed. One of the key points for hardware implementations of DNNs is to reduce their power consumption because DNNs require a lot of product-sum operations. Previous papers presented some accelerators using logarithmic quantization to reduce the power consumption by replacing multipliers with shifters. However, most of them are implemented only for inference. In this paper, an Accelerator of quantization neural networkS with Stochastic approach (AQSS) is proposed. It uses a stochastic approach for logarithmic quantization, and enables DNNs to infer or to learn using logarithmic quantization. A prototype of AQSS is implemented on a field-programmable gate array (FPGA) (Intel Arria 10 GX 1150) and synthesized with Intel Quartus Prime 17.1 Standard Edition. As a result, it is confirmed to have 1.8 times the power efficiency of GPU.

关键词： neural networks deep learning field-programmable gate array hardware logarithmic quantization

来源：评论

学校读者我要写书评

暂无评论

On Stochastic Adaptive Control under quantization 11

On Stochastic Adaptive Control under Quantization

引用

11th World Congress on Intelligent Control and Automation

作者： Tian, Yin Zhang, Hui Zhejiang Univ Dept Control Sci & Engn Natl Key Lab Ind Control Technol Hangzhou 310027 Zhejiang Peoples R China

ISBN: (纸本)9781479958252

This paper discusses the stochastic adaptive tracking problem with input quantization, where the unknown parameters are described as a Gauss-Markov process. Based on the analysis of the cautious control law under quantization, the innovations dual controller is proposed. The quantization errors of logarithmic quantizer introduce the excitation of parameter estimation and hence improve the total closed-loop performance. Inspired by this, we also propose an alternative approach adopting the cautious control law and the logarithmic quantization with preprocessing providing the excitation to guarantee future estimation. The proposed methods imply the exciting effect of the quantization involved in the feedback loop. Simulation illustrates our results.

关键词： stochastic adaptive control logarithmic quantization turn off dual property excitation

来源：评论

学校读者我要写书评

暂无评论

Analysis of Semilogarithmic Companding quantization 14

Analysis of Semilogarithmic Companding Quantization

引用

14th International Conference on Advanced Technologies, Systems and Services in Telecommunications (TELSIKS)

作者： Peric, Zoran H. Jovanovic, Aleksandra Z. Nikolic, Jelena R. Peric, Anastasija Z. Univ Nis Fac Elect Engn Aleksandra Medvedeva 14 Nish 18000 Serbia

ISBN: (纸本)9781728108780

In this paper, an analysis of semilogarithmic companding quantization of amplitudes having the Laplacian probability density function is performed. Particular attention is focused on studying quantizer robustness on changes in amplitude variance. In order to achieve better robustness of the quantizer in question, a method for its design is proposed, which provides the approximately equal quality of the quantized signal at the boundaries of the predefined variance range. The presented analysis and the obtained results can be useful in digital processing of speech and other non-stationary signals having wide amplitude dynamics.

关键词： logarithmic quantization Semilogarithmic compression characteristic Robustness on variance changes Laplacian source

来源：评论

学校读者我要写书评

暂无评论

DISTRIBUTED QUANTIZED CONSENSUS FOR AGENTS ON DIRECTED NETWORKS

引用

Journal of Systems Science & Complexity 2013年第4期26卷 489-511页

作者： LI Dequan LIU Qipeng WANG Xiaofan Department of Automation Shanghai Jiao Tong Universityand Key Laboratory of System Control and Information ProcessingMinistry of Education of China School of Science Anhui University of Science and Technology

Communication bandwidth and network topology are two important factors that affect performance of distributed consensus in multi-agent *** available works about quantized average consensus assume that the adjacency matrices associated with the digraphs are doubly stochastic,which amounts to that the digital networks are ***,this assumption may be unrealistic in *** this paper,without assuming double stochasticity,the authors revisit an existing quantized average consensus protocol with the logarithmic quantization scheme,and investigate the quantized consensus problem in general directed digital networks that are strongly connected but not necessarily *** authors first derive an achievable upper bound of the quantization precision parameter to design suitable logarithmic quantizer,and this bound explicitly depends on network ***,by means of the matrix transformation and the Lyapunov techniques,the authors provide a testable condition under which the weighted average consensus can be achieved with the proposed quantized protocol.

关键词： Consensus protocol digraph logarithmic quantization multi-agent systems.

来源：评论

学校读者我要写书评

暂无评论

Distributed H_∞ fusion filtering with communication bandwidth constraints

引用

SIGNAL PROCESSING 2014年第partb期96卷 284-289页

作者： Chen, Bo Yu, Li Zhang, Wen-An Wang, Hongxia Zhejiang Univ Technol Coll Informat Engn Zhejiang Prov United Key Lab Embedded Syst Hangzhou 310023 Zhejiang Peoples R China

This paper is concerned with the distributed H, fusion filtering problem (DHFFP) for a class of networked multi-sensor fusion systems with communication bandwidth constraints. Due to the limited bandwidth, only finite-level quantized sensor messages are sent to the fusion center, and multiple finite-level logarithmic quantizers are introduced to describe the above quantization strategy. In this sense, the DHFFP is inherent the co-design of the fusion parameters and quantization parameters. With the aid of the discrete-time bounded real lemma, the co-design problem is converted into a convex optimization problem over all the aforementioned parameters, which can be easily solved by standard software packages. It turns out that the performance of the designed distributed fusion filter is superior to that of each local quantized estimate. Finally, a numerical example is given to show the effectiveness of the proposed method. (C) 2013 Elsevier B.V. All rights reserved.

关键词： Distributed H-infinity fusion filter Communication bandwidth constraints logarithmic quantization Convex optimization

来源：评论

学校读者我要写书评

暂无评论

Design of event-based sliding mode controller with logarithmic quantized state measurement and delayed control update

引用

ISA TRANSACTIONS 2022年 124卷 280-289页

作者： Yesmin, Asifa Bera, Manas Kumar Natl Inst Technol Silchar Dept Elect & Instrumentat Engn Silchar 788010 Assam India

This paper concerns with the robust stabilization of a perturbed linear time-invariant (LTI) single-input single-output (SISO) system based on event-based/ event-triggered sliding mode control (SMC) with logarithmic quantized state measurement. This technique is expected to reduce communication traffic in control networks with the assurance of robust stability against the quantization error and external perturbations. In practice, delay is inevitable in the control computation during the implementation of the control law. The proposed event-triggering mechanism is analysed for the two scenarios: with delay and without delay in control computation. The robust stability of the uncertain system is ensured with the external disturbances, the error introduced by the quantization process and delay encountered in control computation with the proposed event-based SMC. The lower bound of minimum inter event execution time is derived to ensure the Zeno free behaviour. To show the effect of delay on the triggering mechanism, the maximum bound of the delay is obtained to guarantee the Zeno free behaviour with the robust stability of the closed loop system. An example of a mechanical system is simulated and results are shown to validate the effectiveness of the proposed methodology. (c) 2020 ISA. Published by Elsevier Ltd. All rights reserved.

关键词： Event-based control Sliding mode control logarithmic quantization Robust stabilization

来源：评论

学校读者我要写书评

暂无评论

Networked Control With Reset Quantized State Based on Bernoulli Processing

引用

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 2014年第9期61卷 4838-4846页

作者： Lu, Renquan Wu, Fang Xue, Anke Hangzhou Dianzi Univ Inst Informat & Control Hangzhou 310018 Zhejiang Peoples R China

This paper considers the problem of reset quantized state control for a class of continuous-time linear system. The logarithmic quantization scheme is employed in the network-based information communication, and a reset state observer is introduced which is based on the standard one to suppress sensor quantization effects. A Bernoulli processing approach is presented to reset the value of the observer with a random way in continuous-time domain. The resulting reset error system is a class of hybrid system which contains both discrete-time and continuous-time stochastic error dynamics. By the stochastic Lyapunov method, it is derived that the error system under the proposed reset technique is asymptotically stable in the sense of each discrete "reset" time instant, and mean-square stable in the continuous-time domain except those "reset" time instants. The observer and controller gains of the closed-loop systems are obtained via solving linear matrix inequalities. Finally, a numerical example is provided to illustrate the effectiveness of the proposed approach.

关键词： Bernoulli processing logarithmic quantization reset state observer-based control

来源：评论

学校读者我要写书评

暂无评论

Consensus tracking via quantized iterative learning control for singular nonlinear multi-agent systems with state time-delay and initial state error

引用

NONLINEAR DYNAMICS 2021年第3期103卷 2701-2719页

作者： Zhou, Xingyu Wang, Haoping Tian, Yang Dai, Xisheng Nanjing Univ Sci & Technol Sch Automat Nanjing 210094 Jiangsu Peoples R China Guangxi Univ Sci & Technol Sch Elect & Informat Engn Liuzhou 545006 Guangxi Peoples R China

This paper investigates the quantized iterative learning consensus tracking problem for singular nonlinear multi-agent systems (MASs) in the presence of state time-delay and initial state error. The unified D-type quantized learning control protocols based on the initial state learning with index gain is proposed and employed for singular nonlinear MASs with state time-delay in both continuous-time domain and discrete-time domain. Based on the operator theory, the convergence condition of the consensus tracking errors between each follower agent and the leader is manifested and analyzed over a fixed time interval. Furthermore, the closed-loop D-type learning protocols are introduced to track the leader's trajectory in order to compare the convergent rate with the variable index gain learning protocols. Finally, simulation results are applied to confirm the validity of the proposed protocols.

关键词： Iterative learning control Singular nonlinear multi-agent systems logarithmic quantization Variable index gain Consensus tracking analysis

来源：评论

学校读者我要写书评

暂无评论

Consensus Control for Directed Networks Under Quantized Information Exchange

引用

ASIAN JOURNAL OF CONTROL 2016年第2期18卷 763-770页

作者： Yin, Zhi-Xiang Huang, You-Rui Geng, Xian-Ya Li, De-Quan Anhui Univ Sci & Technol Sch Sci Huainan 232001 Anhui Peoples R China Anhui Univ Sci & Technol Sch Elect & Informat Engn Huainan 232001 Anhui Peoples R China

In this paper, without assuming balanced network topologies, we address the weighted average consensus problem for discrete-time single-integrator multi-agent systems with logarithmic quantized information communication. By incorporating generalized quadratic Lyapunov function with the discrete-time Bellman-Gronwall inequality, a new upper bound about the quantization precision parameter of the infinite-level logarithmic quantizer is derived to design quantized protocol, under which agents in strongly connected directed networks can attain weighted average consensus. The obtained new upper bound clearly characterizes the intimate relation between the quantization precision parameter and the directed network topology. The proposed quantized protocol is particularly applicable to digital networks where balanced message passing among agents is not available.

关键词： Consensus double stochasticity logarithmic quantization digraph

来源：评论

学校读者我要写书评

暂无评论

Differentially Private Average Consensus With logarithmic Dynamic Encoding-Decoding Scheme

引用

IEEE TRANSACTIONS ON CYBERNETICS 2023年第10期53卷 6725-6736页

作者： Chen, Wei Wang, Zidong Hu, Jun Liu, Guo-Ping Southern Univ Sci & Technol Ctr Control Sci & Technol Shenzhen 518055 Peoples R China Shandong Univ Sci & Technol Coll Elect Engn & Automat Qingdao 266590 Peoples R China Brunel Univ London Dept Comp Sci Uxbridge UB8 3PH England Harbin Univ Sci & Technol Dept Math Harbin 150080 Peoples R China Harbin Univ Sci & Technol Heilongjiang Prov Key Lab Optimizat Control & Inte Harbin 150080 Peoples R China

This article is concerned with the differentially private average consensus (DPAC) problem for a class of multiagent systems with quantized communication. By constructing a pair of auxiliary dynamic equations, a logarithmic dynamic encoding-decoding (LDED) scheme is developed and then utilized during the process of data transmission, thereby eliminating the effect of quantization errors on the consensus accuracy. The primary purpose of this article is to establish a unified framework that integrates the convergence analysis, the accuracy evaluation, and the privacy level for the developed DPAC algorithm under the LDED communication scheme. By means of the matrix eigenvalue analysis method, the Jury stability criterion, and the probability theory, a sufficient condition (with respect to the quantization accuracy, the coupling strength, and the communication topology) is first derived to ensure the almost sure convergence of the proposed DPAC algorithm, and the convergence accuracy and privacy level are thoroughly investigated by resorting to the Chebyshev inequality and $\epsilon$ -differential privacy index. Finally, simulation results are provided to illustrate the correctness and validity of the developed algorithm.

关键词： Differential private average consensus dynamic encoding-decoding scheme logarithmic quantization multiagent systems (MASs)

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：