Description: 一种基于蚁群聚类的径向基神经网络
提出了一种基于蚁群聚类算法的径向基神经网络. 利用蚁群算法的并行寻优特征和挥发系
数方法的自适应更改信息量的能力,并以球面聚类的方式确定了径向基神经网络中基函数的位置,
同时通过比较隐层神经元的相似性、合并相似性较为接近的2 个神经元来约简隐含层的神经元,以
达到简化径向基神经网络结构的目的. 实验比较了几种不同聚类算法的径向基神经网络,结果表
明,所提神经网络的整体训练时间至少可缩短40 ％ ,学习的准确率可提高1 ％以上,而且网络结构
更加精简.-An Ant clustering RBFNN presents a clustering algorithm based on the ant colony RBFN Liaison. Ant use the parallel algorithm optimization features and volatile coefficient method of adaptive changes in the volume of information, spherical cluster and the method of RBF neural network-based functions of the position, By comparing the same time hidden layer neurons of similarity, Merger similarity is closer to the two neurons to Jane about the hidden layer neurons, to achieve simplification RBF neural network structure purposes. Experimental comparison of several different clustering algorithm RBF neural network, The results show that the neural network's overall training time can be shortened at least 40%. Learning the accuracy rate can be increased by 1%, and the network is more streaml Platform: |
Size: 292558 |
Author:zhxj |
Hits:
Description: 这是我参考了n(n>=5,^_^)篇BP神经网优化方法的论文写出的BP神经网源代码，使用了批处理训练方法，带动量项、学习速率的自适应调整、仅对学习精度没有达到指定要求的模式训练，并且训练精度逐步加大，通过这种方法，极大地加快了训练速度；另外，对于0模式和1模式数目相差很大(5-6倍)都能达到很高的学习精度。-This is my reference to the n (NGT; Chapter 5 = ,^_^) BP neural network optimization method to write the thesis of BP neural network source code, use the batch method of training, the volume driven, adaptive learning rate adjustment, not only learning accuracy requirements specified the model training, and training gradually increased accuracy, this method has greatly accelerated the pace of training; In addition, the 0 mode and a model number vary greatly (5-6 times) to achieve high accuracy of the study. Platform: |
Size: 4676 |
Author:王宏志 |
Hits:
Description: 这是我参考了n(n>=5,^_^)篇BP神经网优化方法的论文写出的BP神经网源代码，使用了批处理训练方法，带动量项、学习速率的自适应调整、仅对学习精度没有达到指定要求的模式训练，并且训练精度逐步加大，通过这种方法，极大地加快了训练速度；另外，对于0模式和1模式数目相差很大(5-6倍)都能达到很高的学习精度。-This is my reference to the n (NGT; Chapter 5 = ,^_^) BP neural network optimization method to write the thesis of BP neural network source code, use the batch method of training, the volume driven, adaptive learning rate adjustment, not only learning accuracy requirements specified the model training, and training gradually increased accuracy, this method has greatly accelerated the pace of training; In addition, the 0 mode and a model number vary greatly (5-6 times) to achieve high accuracy of the study. Platform: |
Size: 4096 |
Author:王宏志 |
Hits:
Description: 一种基于蚁群聚类的径向基神经网络
提出了一种基于蚁群聚类算法的径向基神经网络. 利用蚁群算法的并行寻优特征和挥发系
数方法的自适应更改信息量的能力,并以球面聚类的方式确定了径向基神经网络中基函数的位置,
同时通过比较隐层神经元的相似性、合并相似性较为接近的2 个神经元来约简隐含层的神经元,以
达到简化径向基神经网络结构的目的. 实验比较了几种不同聚类算法的径向基神经网络,结果表
明,所提神经网络的整体训练时间至少可缩短40 ％ ,学习的准确率可提高1 ％以上,而且网络结构
更加精简.-An Ant clustering RBFNN presents a clustering algorithm based on the ant colony RBFN Liaison. Ant use the parallel algorithm optimization features and volatile coefficient method of adaptive changes in the volume of information, spherical cluster and the method of RBF neural network-based functions of the position, By comparing the same time hidden layer neurons of similarity, Merger similarity is closer to the two neurons to Jane about the hidden layer neurons, to achieve simplification RBF neural network structure purposes. Experimental comparison of several different clustering algorithm RBF neural network, The results show that the neural network's overall training time can be shortened at least 40%. Learning the accuracy rate can be increased by 1%, and the network is more streaml Platform: |
Size: 291840 |
Author:zhxj |
Hits:
Description: 基于BP神经网络的 参数自学习控制
（1）确定BP网络的结构，即确定输入层节点数M和隐含层节点数Q，并给出各层加权系数的初值 和 ，选定学习速率 和惯性系数 ，此时k=1；
（2）采样得到rin(k)和yout(k)，计算该时刻误差error(k)=rin(k)-yout(k)；
（3）计算神经网络NN各层神经元的输入、输出，NN输出层的输出即为PID控制器的三个可调参数 ， ， ；
（4）根据(3.34)计算PID控制器的输出u(k)；
（5）进行神经网络学习，在线调整加权系数 和 ，实现PID控制参数的自适应调整；
（6）置k=k+1，返回（1）。
-Based on the parameters of BP neural network self-learning control (1) to determine the structure of BP network, that is, determine the input layer nodes M and hidden layer nodes Q, and gives all levels of the initial value and the weighted coefficient, the selected learning rate and inertia coefficient, when k = 1 (2) sample has been rin (k) and the yout (k), calculate the moment of error error (k) = rin (k)-yout (k) (3) calculation of neural network NN all floors of the neurons in input and output, NN output layer is the output of PID controller for the three adjustable parameters,, (4) According to (3.34) Calculation of PID controller output u (k) (5) to carry out neural network learning, on-line adjustment of the weighted coefficient and, realize the adaptive PID control parameters adjust (6) purchase k = k+ 1, return (1). Platform: |
Size: 1024 |
Author:dake |
Hits:
Description: 讨论了BP 小波神经网络在训练过程中减小误差函数时最优方向的确定和自适应调整学习率的方法。
首先论证了小波神经网络的数学基础,然后讨论了BP 小波神经网络的学习过程,重点讨论了减小误差函数最优方
向的确定方法,即如何保证步长方向与负梯度方向一致,由此得出了自适应调整学习率的简便方法。该方法具有
普遍性,有广泛的应用价值。仿真结果表明,采用最优梯度下降方向可以大幅度提高BP 小波神经网络的学习速
度。-Discussed BP wavelet neural network in the training process of reducing the error function to determine the optimal direction and adaptive learning rate adjustment method. First of all, demonstrated that the wavelet neural network mathematical basis, and then discussed the BP wavelet neural network learning process focused on reducing the error function to determine the optimal direction of approach, namely how to ensure that the direction of step with the negative gradient direction, which may out the adaptive learning rate adjustment easy way. The method is universal, has a wide range of application value. Simulation results show that the optimal direction of gradient descent can dramatically improve BP wavelet neural network learning speed. Platform: |
Size: 245760 |
Author:guole |
Hits:
Description: 有关BP神经网络改进算法的一篇论文,增加动量项,自适应学习率等-Of the BP neural network algorithm to improve the paper to increase the momentum, the adaptive learning rate Platform: |
Size: 146432 |
Author:simbabi |
Hits:
Description: 这是一个用动量法和自适应学习速率法改良过后的BP神经网络。编绎环境为VC9.0。当网络训练完成后。能查看误差曲线图。-This is a method with momentum and adaptive learning rate method, after the BP neural network improved. Code Interpretation of the environment VC9.0. When the network training is completed. Can view the error curve. Platform: |
Size: 5518336 |
Author:8085454 |
Hits:
Description: 编写了一个pso优化bp神经网络的程序，应用在分类中。第一步：pso优化bp神经网络得到最优的阈值和权值，第二步bp神经网络把该最优的阈值和权值作为初始阈值和权值，采用动量及自适应学习速率算法进行训练。附件中，是数据和编写的部分程序，tiqushuju是用来提取文本中的数据构造样本集的函数。mubiao是用来构造期望输出的函数。bp是已经编写好的，未使用pso优化的bp神经网络函数。pso是本人编写的pso优化bp神经网络的函数，psobp是采用pso优化的阈值和权值作为bp神经网络的初始权值和阈值进行训练和测试的函数。但是本人编写的粒子群优化bp网络的程序训练效果和测试效果远不如只使用bp的效果。-Prepared a pso bp neural network optimization, application in the classification. First step: PSO optimization bp neural network to get the optimal threshold and weight, the second step bp neural network to the optimal threshold and weight as the initial threshold value and weight, momentum and adaptive learning rate training algorithm . Accessories, data and written part of the program,. Tiqushuju function is used to extract the text data structure of the sample set. mubiao is used to construct the desired output function. BP is already written, unused PSO optimization bp neural network function. pso pso bp neural network optimization function is written in my psobp PSO optimization threshold and the right values as a function of the initial weights and thresholds of BP neural network training and testing. But I am prepared particle swarm optimization BP network training effect and test the effect is far better than using only the effect of the bp. Platform: |
Size: 49152 |
Author:wk |
Hits:
Description: As signatures are widely accepted bio-metric for authentication and identification of a person because every person
has a distinct signature with its specific behavioral property, so it is very much necessary to prove the authenticity of signature
itself. There are various techniques to signature recognition with a lot of scope of research. In this paper off-line signature
recognition and verification system using Artificial Neural Network (ANN) is purposed. The purposed network based upon the
adaption of ANN to recognized signature to connected type pattern. The purpose ANN was trained with back propagation with
momentum and adaptive learning rate. A triple hidden layer ANN with 100 inputs 58.38.20 hidden neurons layers and 5 neurons
in output layers gives best results as compared with other networks. This paper represents a brief review on various approaches
used in signature verification systems. Platform: |
Size: 100352 |
Author:shiwi |
Hits:
Description: 多层神经网络，在训练过程中采用自适应学习率Adagrad方法。可以实现回归或分类问题。(The adaptive learning rate Adagrad method is adopted in the training process of the multilayer neural network. Regression or classification problems can be achieved.) Platform: |
Size: 4096 |
Author:ring0o0o
|
Hits:
Search - adaptive learning rate in neural network