检索结果-内蒙古大学图书馆

An approximate randomization-based neural network with dedicated digital architecture for energy-constrained devices

引用

NEURAL COMPUTING & APPLICATIONS 2023年第9期35卷 6753-6766页

作者： Ragusa, Edoardo Gianoglio, Christian Zunino, Rodolfo Gastaldo, Paolo Univ Genoa Dept Elect Elect Telecommunicat Engn & Naval Architecture DITEN Genoa Italy

Variable energy constraints affect the implementations of neural networks on battery-operated embedded systems. This paper describes a learning algorithm for randomization-based neural networks with hard-limit activation functions. The approach adopts a novel cost function that balances accuracy and network complexity during training. From an energy-specific perspective, the new learning strategy allows to adjust, dynamically and in real time, the number of operations during the network's forward phase. The proposed learning scheme leads to efficient predictors supported by digital architectures. The resulting digital architecture can switch to approximate computing at run time, in compliance with the available energy budget. Experiments on 10 real-world prediction testbeds confirmed the effectiveness of the learning scheme. Additional tests on limited-resource devices supported the implementation efficiency of the overall design approach.

关键词： randomization-based neural networks Threshold function Low-power Digital architectures FPGA Extreme learning machine random vector functional link networks

来源：评论

学校读者我要写书评

暂无评论

Letter on Convergence of In-Parameter-Linear Nonlinear Neural Architectures With Gradient Learnings

引用

IEEE TRANSACTIONS ON NEURAL networks AND LEARNING SYSTEMS 2023年第8期34卷 5189-5192页

作者： Bukovsky, Ivo Dohnal, Gejza Benes, Peter M. Ichiji, Kei Homma, Noriyasu Univ South Bohemia Ceske Budejovice Dept Comp Sci Fac Sci Ceske Budejovice 37005 Czech Republic Czech Tech Univ Fac Mech Engn Dept Mech Biomechan & Mechatron Prague 16607 Czech Republic Czech Tech Univ Dept Tech Math Fac Mech Engn Prague 16607 Czech Republic Siemens Mobil Sro Dept Rail Automat Prague 18200 Czech Republic Tohoku Univ Grad Sch Med Dept Radiol Imaging & Informat Sendai Miyagi 9808575 Japan

This letter summarizes and proves the concept of bounded-input bounded-state (BIBS) stability for weight convergence of a broad family of in-parameter-linear nonlinear neural architectures (IPLNAs) as it generally applies to a broad family of incremental gradient learning algorithms. A practical BIBS convergence condition results from the derived proofs for every individual learning point or batches for real-time applications.

关键词： Convergence Stability criteria Neural networks Mechanical engineering Learning systems Loss measurement Kernel Bounded-input bounded-state stability (BIBS) extreme learning machines in-parameter-linear nonlinear neural architectures (IPLNAs) incremental gradient learnings input-to-state stability (ISS) polynomial neural networks random vector functional link networks weight convergence

来源：评论

学校读者我要写书评

暂无评论

Electricity Load Demand Time Series Forecasting with Empirical Mode Decomposition based random vector functional link Network

Electricity Load Demand Time Series Forecasting with Empiric...

引用

IEEE International Conference on Systems, Man, and Cybernetics

作者： Xueheng Qiu P. N. Suganthan Gehan A. J. Amaratunga School of Electrical and Electronic Engineering Nanyang Technological Univeristy Singapore Department of Engineering University of Cambridge UK

ISBN: (纸本)9781509018987

Short-term electricity load demand forecasting is a critical process in the management of modern power system. An ensemble method composed of Empirical Mode Decomposition (EMD) and random vector functional link network (RVFL) is presented in this paper. Due to the randomly generated weights between input and hidden layers and the close form solution for parameter tuning, RVFL network is a universal approximator with the advantages of fast training. By introducing ensemble approach via EMD into RVFL network, the performance can be significantly improved. Five electricity load demand datasets from Australian Energy Market Operator (AEMO) were used to evaluate the performance of the proposed method. The attractiveness of the proposed EMD based RVFL network can be demonstrated by the comparison with six benchmark methods.

关键词： Support vector regression Ensemble learning Load demand time series forecasting random vector functional link networks Neural networks

来源：评论

学校读者我要写书评

暂无评论

Gauss–Seidel Extreme Learning Machines

引用

SN Computer Science 2020年第4期1卷 1-28页

作者： de Freitas, Rafaela C. Ferreira, Janderson de Lima, Sidney M. L. Fernandes, Bruno José T. Bezerra, Byron L. D. dos Santos, Wellington P. Polytechnique School of the University of Pernambuco Recife Brazil Federal University of Pernambuco Recife Brazil

Extreme learning machines (ELM) were created to simplify the training phase of single-layer feedforward neural networks, where the input weights are randomly set and the only parameter is the number of neurons in the hidden layer. These networks are also known for one-shot training using Moore–Penrose pseudo-inverse. In this work, we propose Gauss–Seidel extreme learning machine (GS-ELM), an ELM based on Gauss–Seidel iterative method to solve linear equation systems. We performed tests considering databases with different characteristics and analysed its discrimination capabilities and memory consumption in comparison to the canonical ELM and the online sequential ELM. GS-ELM presented similar discrimination capabilities, but consuming significantly less memory, turning possible its application in low-memory systems and embedded solutions. © 2020, Springer Nature Singapore Pte Ltd.

关键词： Extreme learning machines Fast training neural networks Numerical methods random vector functional link networks Unorganized neural networks

来源：评论

学校读者我要写书评

暂无评论

Generalized Learning vector Quantization for Classification in randomized Neural networks and Hyperdimensional Computing

Generalized Learning Vector Quantization for Classification ...

引用

International Joint Conference on Neural networks (IJCNN)

作者： Diao, Cameron Kleyko, Denis Rabaey, Jan M. Olshausen, Bruno A. Rice Univ Dept Comp Sci Houston TX 77005 USA Univ Calif Berkeley Berkeley CA USA Res Inst Sweden Kista Sweden Univ Calif Berkeley Berkeley Wireless Res Ctr Berkeley CA USA Univ Calif Berkeley Redwood Ctr Theoret Neurosci Berkeley CA USA

ISBN: (纸本)9780738133669

Machine learning algorithms deployed on edge devices must meet certain resource constraints and efficiency requirements. random vector functional link (RVFL) networks are favored for such applications due to their simple design and training efficiency. We propose a modified RVFL network that avoids computationally expensive matrix operations during training, thus expanding the network's range of potential applications. Our modification replaces the least-squares classifier with the Generalized Learning vector Quantization (GLVQ) classifier, which only employs simple vector and distance calculations. The GLVQ classifier can also be considered an improvement upon certain classification algorithms popularly used in the area of Hyperdimensional Computing. The proposed approach achieved state-of-the-art accuracy on a collection of datasets from the UCI Machine Learning Repository-higher than previously proposed RVFL networks. We further demonstrate that our approach still achieves high accuracy while severely limited in training iterations (using on average only 21% of the least-squares classifier computational costs).

关键词： learning vector quantization randomly connected neural networks hyperdimensional computing random vector functional link networks

来源：评论

学校读者我要写书评

暂无评论

Hyperdimensional Computing for Efficient Distributed Classification with randomized Neural networks

Hyperdimensional Computing for Efficient Distributed Classif...

引用

International Joint Conference on Neural networks (IJCNN)

作者： Rosato, Antonello Panella, Massimo Kleyko, Denis Univ Roma La Sapienza Diet Dept Rome Italy Univ Calif Berkeley Berkeley CA USA Res Inst Sweden Kista Sweden

ISBN: (纸本)9780738133669

In the supervised learning domain, considering the recent prevalence of algorithms with high computational cost, the attention is steering towards simpler, lighter, and less computationally extensive training and inference approaches. In particular, randomized algorithms are currently having a resurgence, given their generalized elementary approach. By using randomized neural networks, we study distributed classification, which can be employed in situations were data cannot be stored at a central location nor shared. We propose a more efficient solution for distributed classification by making use of a lossy compression approach applied when sharing the local classifiers with other agents. This approach originates from the framework of hyperdimensional computing, and is adapted herein. The results of experiments on a collection of datasets demonstrate that the proposed approach has usually higher accuracy than local classifiers and getting close to the benchmark - the centralized classifier. This work can be considered as the first step towards analyzing the variegated horizon of distributed randomized neural networks.

关键词： randomized neural networks hyperdimensional computing random vector functional link networks

来源：评论

学校读者我要写书评

暂无评论

Broad Learning System: structural extensions on single-layer and multi-layer neural networks

Broad Learning System: structural extensions on single-layer...

引用

International Conference on Security, Pattern Analysis, and Cybernetics (ICSPAC)

作者： Liu, Zhulin Chen, C. L. Philip Univ Macau Fac Sci & Technol Dept Comp & Informat Sci Macau Peoples R China Dalian Maritime Univ Dalian Peoples R China Chinese Acad Sci Inst Automat State Key Lab Management & Control Complex Syst Beijing Peoples R China

ISBN: (纸本)9781538630167

Broad Learning System proposed recently Ill demonstrates efficient and effective learning capability. Moreover, fast incremental learning algorithms are developed in broad expansions without an entire retraining of the whole model. Compared with the systems in deep structure, the inspired system provides competitive results in classification. In this paper, the broad learning algorithms and incremental learning algorithms are applied to commonly used neural networks, such as radial basis function neural networks (RBF) and hierarchical extremal learning machine (H-ELM). For RBF, the resulting models, called BLS-RBF, are established by regarding the radial basis function as the mapping in the enhancement nodes, and additional enhancement nodes are added if the network needs expansion widely. For H-ELM, the established model, is developed for the incremental extension of multilayer structure. The developed BLS models and algorithms are very effective and efficient in classification. Finally, experimental results are presented.

关键词： Single layer feedforward neural networks random vector functional link networks broad learning system incremental learning extremal learning machine radial basis function networks hierarchical extremal learning machine

来源：评论

学校读者我要写书评

暂无评论

Impact of Probability Distribution Selection on RVFL Performance 1

引用

2nd International Conference on Smart Computing and Communication (SmartCom)

作者： Cao, Weipeng Gao, Jinzhu Ming, Zhong Cai, Shubin Zheng, Hua Shenzhen Univ Shenzhen 518060 Peoples R China Univ Pacific Stockton CA 95211 USA

ISBN: (数字)9783319738307

ISBN: (纸本)9783319738307;9783319738291

The initialization of input weights and hidden biases plays an important role in random vector functional link networks (RVFL). Although some optimization algorithms for initialization have been proposed in recent years, the initialization strategies of these algorithms are under the premise of the uniform distribution. In this paper, ten benchmark datasets are used to study the impact of different probability distributions (e.g., Uniform, Gaussian, and Gamma distributions) initialization on the performance of RVFL. The experimental results present some interesting observations and valuable instructions: (1) No matter whether we use Uniform, Gaussian, or Gamma distributions, RVFL initialized by the distribution with smaller variances always get lower training and testing RMSE;(2) Compared with the Uniform distribution, the Gaussian and Gamma distributions with smaller variances usually give the RVFL model better performance;(3) Regardless of the distribution, RVFL with the direct link from the input layer to the output layer has better performance than those without the link;(4) RVFL initialized by the distribution with larger variances generally needs more hidden nodes to achieve equivalent accuracy with ones having the smaller variances;(5) With the increase of distribution variances, the performance of RVFL decreases first and then remains stable.

关键词： random vector functional link networks Neural networks with random weights random initialization Probability distribution Parameter selection

来源：评论

学校读者我要写书评

暂无评论

Broad Learning System: Feature extraction based on K-means clustering algorithm

Broad Learning System: Feature extraction based on K-means c...

引用

4th International Conference on Information, Cybernetics and Computational Social Systems (ICCSS)

作者： Liu, Zhulin Zhou, Jin Chen, C. L. Philip Univ Macau Dept Comp & Informat Sci Fac Sci & Technol Macau Peoples R China Dalian Maritime Univ Dalian Peoples R China Chinese Acad Sci Inst Automat State Key Lab Management & Control Complex Syst Beijing Peoples R China Univ Jinan Shandong Prov Key Lab Network Based Intelligent C Jinan Shandong Peoples R China

ISBN: (纸本)9781538632574

Broad Learning System Hi proposed recently demonstrates efficient and effective learning capability. This model is also proved to be suitable for incremental learning algorithms by taking the advantages of random vector flat neural networks. In this paper, a modified BLS structure based on the K-means feature extraction is developed. Compared with the original broad learning system, acceptable performance on more complicated data set, such as CIFAR-10, is achieved. Furthermore, it is proved that the proposed model in Hi is flexible and potential in various applications.

关键词： Single layer feedforward neural networks SVD random vector functional link networks broad learning system incremental learning deep learning K-means feature representation

来源：评论

学校读者我要写书评

暂无评论

Broad Learning System: a new learning paradigm and system without going deep 32

Broad Learning System: a new learning paradigm and system wi...

引用

32nd Youth Academic Annual Conference of Chinese Association of Automation (YAC)

作者： Chen, C. L. Philip Liu, Zhulin Univ Macau Fac Sci & Technol Macau 99999 Peoples R China Dalian Maritime Univ Dalian Liaoning Peoples R China

ISBN: (纸本)9781538629017

This paper introduces a Broad Learning System that gives a new paradigm and learning system without the need of deep architecture. In deep structure and learning, the abundant connecting parameters in filters and layers lead to a time-consuming training process. Broad Learning system, which is established as a flat network, maps the original inputs as mapped features in feature nodes and the structure is expanded in wide sense in the enhancement nodes. Model construction and learning algorithms are introduced here. Moreover, different approaches for the construction of enhancement nodes are given. The advantage of the Broad Learning System is that the learning can be updated dynamically and incrementally without going through a retraining process if the model deems to be expanded on additional feature nodes and enhancement nodes such that the learning is so efficient and effective. In addition, The incremental learning algorithms can be conveniently implemented for fast remodeling in broad expansion which can be referred in [1]. Compared with existing deep neural networks, experimental results on the MNIST data manifest the effectiveness of the Broad Learning System.

关键词： Single layer feedforward neural networks random vector functional link networks broad learning system deep learning

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：