检索结果-内蒙古大学图书馆

4th International Conference on Optimization and Learning (OLA)

作者： Fischetti, Matteo Stringher, Matteo Univ Padua Dept Informat Engn Via Gradenigo 6-A I-35100 Padua Italy

ISBN: (纸本)9783030856717;9783030856724

We propose a new metaheuristic training scheme forMachine Learning that combines Stochastic Gradient Descent (SGD) and Discrete Optimization in an unconventional way. Our idea is to define a discrete neighborhood of the current SGD point containing a number of "potentially good moves" that exploit gradient information, and to search this neighborhood by using a classical metaheuristic scheme borrowed from Discrete Optimization. In the present paper we investigate the use of a simple Simulated Annealing (SA) metaheuristic that accepts/rejects a candidate new solution in the neighborhood with a probability that depends both on the new solution quality and on a parameter (the temperature) which is modified over time to lower the probability of accepting worsening moves. Computational results on image classification (CIFAR-10) are reported, showing that the proposed approach leads to an improvement of the final validation accuracy for modern Deep Neural Networks such as ResNet34 and VGG16.

关键词： Simulated annealing Stochastic gradient descent Deep neural networks Machine learning training algorithm

来源：评论

学校读者我要写书评

暂无评论

Dynamic Feature Extraction of Power Disturbance Signal Based on Time-frequency Technology

Dynamic Feature Extraction of Power Disturbance Signal Based...

引用

21st Chinese Control and Decision Conference

作者： Wang Yuguo Zhao Wei Xie Yan Hebei Univ Engn Handan 056038 Peoples R China

ISBN: (纸本)9781424427239

The growing concern for power quality issues from both utilities and power users is generated by proliferation of power electronic devices and nonlinear loads in power system network. Therefore, the techniques for power quality monitoring and power disturbance mitigation are capturing increasing attention. A novel approach for the power quality disturbances recognition using wavelet transform and neural network is proposed. The wavelet transform is used to complete feature extraction and can accurately localizes the characteristics of transient signal both in time and frequency domains. These feature vectors are input variables for neural network training and the neural network structure is designed for disturbance pattern recognition. Therefore, the wavelet network combines advantages of wavelet transformation for purposes of feature extraction and selection with the characteristic decision capabilities of neural network approaches. During the training process, the wavelet network learns adequate decision functions and arbitrarily complex decision regions defined by the weight coefficients. The simulation results demonstrate the proposed method gives a new way for signal analysis and pattern recognition of power quality disturbances.

关键词： Power quality disturbance wavelet transform feature extraction neural network pattern recognition training algorithm

来源：评论

学校读者我要写书评

暂无评论

An Intuitionistic Fuzzy Neural Network with Triangular Membership Function

An Intuitionistic Fuzzy Neural Network with Triangular Membe...

引用

Chinese Intelligent Automation Conference (CIAC)

作者： Zhou, Xiaoguang Zhao, Renhou Zhang, Lili Flight Simulat & Training CenterNavy Aviat Acad Hu Ludao 125001 Liaoning Peoples R China Changchun Univ Sci & Technol Coll Optoelect Engn Changchun 130022 Jilin Peoples R China

ISBN: (纸本)9783642385247;9783642385230

In this paper, an intuitionistic fuzzy neural network model with a triangular membership and 1 minus triangular form non-membership functions is proposed. The network structure has six layers, and adopts Mandani's fuzzy reasoning. A new fuzzy inference system is applied in the model, which contains hesitation margin as a part. The two steps dynamic optimal training algorithms for the IFNN is development, the first step is membership function parameters training, the second step is membership function parameters training, which can promise the summing of the trained membership and non-membership functions on finite universal set less then 1. An example is given to demonstrate the intuitionistic fuzzy neural network has a higher degree of accuracy and higher learning efficiency than fuzzy neural network.

关键词： Intuitionistic fuzzy neural networks Fuzzy inference system training algorithm

来源：评论

学校读者我要写书评

暂无评论

Numerical and Neural Network Modeling of Motors of a Robot 3

Numerical and Neural Network Modeling of Motors of a Robot

引用

3rd RSI/ISM International Conference on Robotics and Mechatronics (ICROM)

作者： Tourajizadeh, H. Manteghi, S. Nekoo, S. R. Kharazmi Univ Fac Engn Dept Mech Engn Tehran Iran Islamic Azad Univ Coll Comp & Mechatron Engn Branch Sci & Res Tehran Iran Iran Univ Sci & Technol Sch Mech Engn Tehran Iran

ISBN: (纸本)9781467372343

In this paper, parametric and numerical model of the motors of a robot are extracted. A method is proposed here to control the torque and velocity of the motor simultaneously using the extracted dynamics of the motor and consequently control the robot motion more accurately. Parametric model of the motors are derived by conducting standard tests like locked rotor test and step and sine wave input test. In order to derive the neural network and numerical models, a set of sinusoidal, triangular, and random steps signal, are applied as the input to the motor and its speed is recorded as the output. Neural network model of the motors is extracted by using these dataset and considering the MLP neural network structure with Levenberg _Marquardt training method. Results of the numerical model and parametric models are compared and validated by experimental tests.

关键词： Robot Modeling of DC motor Torque-speed control training algorithm Levenberg-Marquardt

来源：评论

学校读者我要写书评

暂无评论

An experience in image compression using neural networks

An experience in image compression using neural networks

引用

48th International Symposium ELMAR-2006 on Multimedia Signal Processing and Communications

作者： Vilovic, Ivan Univ Dubrovnik Dubrovnik Croatia

ISBN: (纸本)9789537044039

There are number trials of using neural networks as signal processing tools for image compression. In this paper, a direct solution method is used for image compression using the neural networks. An experience of using multilayer perceptron for image compression is presented. The multilayer perceptron is used for transform coding of the image. The network is trained for different number of hidden neurons with direct impact to compress ratio. It is experimented with different images that have been segmented in the blocks of various sizes for compression process. Reconstructed image is compared with original image using signal-to-noise ratio and number of bits per pixel. The results show the possibility of using multilayer perceptrons for image compression.

关键词： image compression neural networks multilayer perceptron training algorithm NMSE SNR

来源：评论

学校读者我要写书评

暂无评论

Fingerprint matching using rotational invariant image based descriptor and machine learning techniques

Fingerprint matching using rotational invariant image based ...

引用

6th International Conference on Emerging Trends in Engineering and Technology (ICETET)

作者： Kumar, Ravinder Chandra, Pravin Hanmandlu, Madasu GGSIP Univ Univ Sch Informat & Comm Technol Sect 16C Delhi India Indian Inst Technol Dept Elect Engn New Delhi 110016 India

ISBN: (纸本)9781479925605

The reliability of fingerprint matching system is highly depends on the perfect alignment algorithm and a suitable matching techniques, which assign a label to the input fingerprint image. In this paper, we propose a rotation invariant fingerprint descriptor and a improved generalization performance classifier. The proposed new descriptor is represented by a histogram of local directional pattern (LDP) computed from extracted region of interest (ROI) of fingerprint images. For fingerprint matching, we propose a single hidden layer neural network (SLFN), which combines a powerful extreme learning machine (ELM) and a well generalized resilient propagation (RPROP) algorithm. The proposed fingerprint matching system comprises the following steps: fingerprint pre-processing / enhancement, ROI extraction, invariant LDP feature extraction, and matching using proposed hybrid classifier. The experimental result shows that the matching accuracy of the proposed system is improved as compare to ELM for lower values of hidden nodes, and other distance based matching approaches proposed in the literature.

关键词： SLFN training algorithm local directional pattern fingerprint matching region of interest (ROI) ELM hybrid RP ELM

来源：评论

学校读者我要写书评

暂无评论

GLOBALLY CONVERGENT MULTILEVEL training OF DEEP RESIDUAL NETWORKS

引用

SIAM JOURNAL ON SCIENTIFIC COMPUTING 2023年第3期45卷 S254-S280页

作者： Kopanicakova, Alena Krause, Rolf Univ Svizzera italiana Euler Inst Lugano Switzerland

We propose a globally convergent multilevel training method for deep residual networks (ResNets). The devised method can be seen as a novel variant of the recursive multilevel trustregion (RMTR) method, which operates in hybrid (stochastic-deterministic) settings by adaptively adjusting minibatch sizes during the training. The multilevel hierarchy and the transfer operators are constructed by exploiting a dynamical system's viewpoint, which interprets forward propagation through the ResNet as a forward Euler discretization of an initial value problem. In contrast to traditional training approaches, our novel RMTR method also incorporates curvature information on all levels of the multilevel hierarchy by means of the limited-memory SR1 method. The overall performance and the convergence properties of the our multilevel training method are numerically investigated using examples from the field of classification and regression.

关键词： trustregion methods multilevel minimization deep residual networks training algorithm

来源：评论

学校读者我要写书评

暂无评论

Weighted generative adversarial network for many-to-many voice conversion 23

Weighted generative adversarial network for many-to-many voi...

引用

23rd International Congress on Acoustics: Integrating 4th EAA Euroregio, ICA 2019

作者： Paul, Dipjyoti Pantazis, Yannis Stylianou, Yannis University of Crete Greece Institute of Applied and Computational Mathematics FORTH Greece

ISBN: (纸本)9783939296157

The goal of voice conversion (VC) is to convert speech from a source speaker to that of a target, without changing phonetic contents. VC usually relies on parallel data for training, which limits its practical applications. Existing approaches are also limited in handling multiple speakers, since different models should be built independently for every speaker pair. To tackle that, a variant of Generative Adversarial Network (StarGAN-VC) were introduced that allows many-to-many mapping instead of learning all the pairwise transformations. Moreover, StarGAN-VC can handle non-parallel data, i.e., speakers do not need to utter the same sentences. In this paper, we suggest an algorithmic variation of StarGAN training where suitable weights are introduced. Weights which modify the Generator's gradient value aim to put more power to fake samples that fool the Discriminator. The suggested algorithm results in a stronger Generator. We refer to this variation as weighted-StarGAN (weStarGAN). In weStarGAN, the convergence of the training performance is accelerated. More importantly, the proposed algorithm achieves significant improvement against baseline StarGAN-VC concerning speech subjective quality for both speech quality and speaker similarity. © 2019 Proceedings of the International Congress on Acoustics. All rights reserved.

关键词： Generative adversarial networks training algorithm Voice conversion

来源：评论

学校读者我要写书评

暂无评论

Analysis of training parameters in the ANN learning process to mapping the concrete carbonation depth

引用

Journal of Building Pathology and Rehabilitation 2019年第1期4卷 16-16页

作者： Felix, Emerson Felipe Possan, Edna Carrazedo, Rogério School of Engineering University of São Paulo (USP) Av. Trabalhador São-Carlense 400 - Parque Arnold Schimidt São Carlos 13566-590 SP Brazil Engineering Department Federal University of Latin American Integration (UNILA) Av. Tancredo Neves 5057 Casa 35 Foz do Iguaçu 85567-000 Brazil

We propose an artificial neural network (ANN) model to predict the CO2 diffusion through the concrete to determine the carbonation depth over time, analyzing the influence of some training algorithm and the network ar... 详细信息

We propose an artificial neural network (ANN) model to predict the CO₂ diffusion through the concrete to determine the carbonation depth over time, analyzing the influence of some training algorithm and the network architecture in the ANN learning process. A reliable experimental test database of the non-accelerated test with 278 results of concrete carbonation depth was created from the published literature. It was used to train, test, and validate the model. Altogether, 120 networks had been trained with different characteristics, verifying its performance. In spite of the non-linearity and complexity of the concrete carbonation phenomenon, the proposed ANN model yielded accurate prediction. Results indicate the best training algorithm and the optimum number of neurons in the hidden layer that allows faster ANN training process and generates the most accurate mapping for the concrete carbonation phenomenon. The use of ANN appears as a robust tool easily applied to the study of the concrete carbonation, aiding in decision making in engineering projects focused on durability. © 2019, Springer Nature Switzerland AG.

关键词： Artificial neural network CO2 diffusion Composite training algorithm

来源：评论

学校读者我要写书评

暂无评论

A fuzzy recurrent artificial neural network (FRANN) for pattern classification

引用

INTERNATIONAL JOURNAL OF UNCERTAINTY FUZZINESS AND KNOWLEDGE-BASED SYSTEMS 2000年第5期8卷 525-538页

作者： Brouwer, RK Univ Coll Cariboo Dept Comp Sci Kamloops BC Canada

This paper proposes a recurrent neural network of fuzzy units, which may be used for approximating a hetero-associative mapping and also for pattern classification. Since classification is concerned with set membership, and objects generally belong to sets to various degrees, a fuzzy network seems a natural for doing classification. In the network proposed here each fuzzy unit defines a fuzzy set. The fuzzy unit in the network determines the degree to which the input vector to the unit lies in that fuzzy set. The fuzzy unit may be compared to a perceptron in which case the input vector is compared to the weighting vector associated with the unit by taking the dot product. The resulting membership value in case of the fuzzy unit is compared to a threshold. training of a fuzzy unit is based on an algorithm for solving linear inequalities similar to the method used for Ho-Kashyap recording. training of the whole network is done by training each unit separately. The training algorithm is tested by trying the algorithm out on representations of letters of the alphabet with their noisy versions. The results obtained by the simulation are very promising.

关键词： fuzzy recurrent neural networks fuzzy sets training algorithm fuzzy perceptron attractor network fuzzy neurons

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：