多输出神经元模型的多层前向神经网络及其应用

研究了一种新的多输出神经元模型.首先,给出这类模型的一般形式,并将该模型应用于多层前向神经网络;其次,给出了其学习算法,即递推最小二乘算法,最后通过几个模拟实验表明,采用多输出神经元模型的多层前向神经网络,具有结构简单,泛化能力强,收敛速度快,收敛精度高等特点,其性能远远优于激活函数可调模型的多层前向神经网络.     

Dynamics of a neuron exposed to integer- and fractional-order discontinuous external magnetic flux

We propose a modified Fitzhugh-Nagumo neuron (MFNN) model. Based on this model, an integer-order MFNN system (case A) and a fractional-order MFNN system (case B) were investigated. In the presence of electromagnetic induction and radiation, memductance and induction can show a variety of distributions. Fractional-order magnetic flux can then be considered. Indeed, a fractional-order setting can be acceptable for non-uniform diffusion. In the case of an MFNN system with integer-order discontinuous magnetic flux, the system has chaotic and non-chaotic attractors. Dynamical analysis of the system shows the birth and death of period doubling, which is a sign of antimonotonicity. Such a behavior has not been studied previously in the dynamics of neurons. In an MFNN system with fractional-order discontinuous magnetic flux, different attractors such as chaotic and periodic attractors can be observed. However, there is no sign of antimonotonicity.

     

Time series prediction with single multiplicative neuron model

Single neuron models are typical functional replica of the biological neuron that are derived using their individual and group responses in networks. In recent past, a lot of work in this area has produced advanced neuron models for both analog and binary data patterns. Popular among these are the higher-order neurons, fuzzy neurons and other polynomial neurons. In this paper, we propose a new neuron model based on a polynomial architecture. Instead of considering all the higher-order terms, a simple aggregation function is used. The aggregation function is considered as a product of linear functions in different dimensions of the space. The functional mapping capability of the proposed neuron model is demonstrated through some well known time series prediction problems and is compared with the standard multilayer neural network.     

Neural network learning with generalized-mean based neuron model

The advances in biophysics of computations and neurocomputing models have brought the foreground importance of dendritic structure of neuron. These structures are assumed as basic computational units of the neuron, capable of realizing the various mathematical operations. The well structured higher order neurons have shown improved computational power and generalization ability. However, these models are difficult to train because of a combinatorial explosion of higher order terms as the number of inputs to the neuron increases. In this paper we present a neural network using new neuron architecture i.e., generalized mean neuron (GMN) model. This neuron model consists of an aggregation function which is based on the generalized mean of all the inputs applied to it. The resulting neuron model has the same number of parameters with improved computational power as the existing multilayer perceptron (MLP) model. The capability of this model has been tested on the classification and time series prediction problems.     

基于泛函分析的人工神经元模型改进研究

人工神经网络是对生物神经系统的模拟,它的信息处理功能与网络单元的特性密切相关。神经元的研究方兴未艾,神经元的效率问题有待提升。通过研究神经元学习算法,提高神经元的学习效率,并且提出一种新的神经元模型。     

神经元网络模型的弱信号随机共振检测研究

基于FitzHugh-Nagumo可兴奋细胞耦合后形成的神经元网络模型,对生物神经系统的弱周期信号随机共振检测机制进行研究。以加和网络的双层FHN神经元模型为例,对周期随机共振现象分别进行研究,并应用信噪比、互信息率对比评价方法,结合输出神经元动作电位的发放频率和幅值,从多个角度进行了定量和定性的描述和比较。实验结果表明,双层FHN神经元网络的随机共振响应优于单神经元的FHN模型,且具有更好的稳定性,可以在一定的噪声强度范围内对输入信号进行有效地检测。     

Multiplicative neuron model artificial neural network based on Gaussian activation function

Multiplicative neuron model-based artificial neural networks are one of the artificial neural network types which have been proposed recently and have produced successful forecasting results. Sigmoid activation function was used in multiplicative neuron model-based artificial neural networks in the previous studies. Although artificial neural networks which involve the use of radial basis activation function produce more successful forecasting results, Gaussian activation function has not been used for multiplicative neuron model yet. In this study, rather than using a sigmoid activation function, Gaussian activation function was used in multiplicative neuron model artificial neural network. The weights of artificial neural network and parameters of activation functions were optimized by guaranteed convergence particle swarm optimization. Two major contributions of this study are as follows: the use of Gaussian activation function in multiplicative neuron model for the first time and the optimizing of central and propagation parameters of activation function with the weights of artificial neural network in a single optimization process. The superior forecasting performance of the proposed Gaussian activation function-based multiplicative neuron model artificial neural network was proved by applying it to real-life time series.     

一种量子神经网络模型学习算法及应用

提出一种量子神经网络模型及学习算法. 首先基于生物神经元信息处理机制和量子计算原理构造出一种量子神经元, 该神经元由加权、聚合、活化、激励四部分组成. 然后由量子神经元构造出三层量子神经网络模型, 其输入和输出为实值向量, 权值和活性值为量子比特. 基于梯度下降法构造了该模型的超线性收敛学习算法. 通过模式识别和函数逼近两种仿真结果表明该模型及算法是有效的.     

基于TS模型的增益自校正单神经元PID控制算法

将单神经元自适应PID控制算法和基于TS模型的模糊推理系统相结合,提出了增益模糊自校正单神经元PID控制算法。该算法使得传统单神经元自适应PID控制的神经元增益具备在线自动调整的功能,对于变增益的不确定性系统的控制,获得了很好的控制性能。     

离散时间Hopfield网络的动力系统分析

离散时间的Hopfield网络模型是一个非线性动力系统．对网络的状态变量引入新的能量函数，利用凸函数次梯度性质可以得到网络状态能量单调减少的条件．对于神经元的连接权值且激活函数单调非减(不一定严格单调增加)的Hopfield网络，若神经元激活函数的增益大于权值矩阵的最小特征值，则全并行时渐进收敛；而当网络串行时，只要网络中每个神经元激活函数的增益与该神经元的自反馈连接权值的和大于零即可．同时，若神经元激活函数单调，网络连接权值对称，利用凸函数次梯度的性质，证明了离散时间的Hopfield网络模型全并行时收敛到周期不大于2的极限环．     

智能神经元网络时间序列预测模型的研究

人工神经元网络的研究技术在理论和实际应用上已经比较成熟,在信号处理系统中也采用该技术进行非线性时间序列信号的预测分析。但是由于该理论黑箱模型的特点,无法引入先验知识,从而预测精度难以提高。针对该问题,文中提出了智能神经网络的动态预测模型,引入智能神经元,建立区别于传统神经网络的预测模型,达到了较为理想的预测效果。并以工业生产参数的时间序列预测——某油井生产过程中MinCurrent参数值,作为实验模型,对该方法进行了验证,结果表明了该模型预测精度较高、计算速度快。     

基于三次样条函数拟合的过程神经元网络训练

过程神经元网络的提出为大样本识别问题开辟了新途径，但其训练方法目前主要基于权函数正交基展开。这种方法基函数个数选取目前尚无理论依据。提出了基于三次样条函数拟合的过程神经元网络训练方法。首先将样本函数、过程神经元权函数的离散化数据拟合成分段表示的三次样条函数，然后计算样本样条函数与权值样条函数乘积在给定采样区间上的积分，并将此积分值提交给网络的过程隐层神经元，输出层由普通神经元组成。三次样条函数具有很好的光滑性、可积性、阶数低、参数少等优点，有效地简化了网络的时空聚合运算。实验表明该方法是可行的。     

一种定点数权值神经网络的优化方法及其应用

为了使得神经网络的应用符合嵌入式系统快速计算、存储量精简的要求,提出了一种定点数权值神经网络的优化方法。采用精度可调的比例数格式定点数表示神经网络的权值和阈值,用遗传算法对神经网络进行训练,并用最小二乘法对网络的非线性连续节点激励函数进行了线性离散化。将这种优化的神经网络应用于触摸屏校准。实验表明,采用该方法进行触摸屏校准比传统的校准方法具有更高的准确率。     

广义神经网络的研究及其在交通流预测中的应用

提出一种基于线性独立函数的智能神经元模型,理论分析表明这种智能神经元比普通神经元具有更高的知识存储能力,可使整个神经网络的信息处理能力得到大大增强.将这种智能神经元组成的广义神经网络应用于交通流预测,对大连市实测交通流数据进行了预测分析.实验结果表明,该广义神经网络的预测效果远远优于常规BP网络,具有较高的实用价值.     

神经元环路中主神经元的模型分析

神经元环路是大脑神经系统的基本单元,而环路的信息输出则由主神经元所决定.本文通过对小脑皮层、嗅球和海马CA1三类不同环路中的主神经元：浦肯野神经元、僧帽神经元和锥体神经元建立给予神经元几何形态和电缆传递的多房室模型,通过它们的动作电位分析和比较,说明不同类型的外界刺激,三个环路中主神经元动作电位的区别;并进一步比较对同类刺激,三类主神经元动作电位的不同,模型验证它们所属环路在神经系统中所起的不同输出功能.     

The hysteretic Hopfield neural network

A new neuron activation function based on a property found in physical systems-hysteresis-is proposed. We incorporate this neuron activation in a fully connected dynamical system to form the hysteretic Hopfield neural network (HHNN). We then present an analog implementation of this architecture and its associated dynamical equation and energy function. We proceed to prove Lyapunov stability for this new model, and then solve a combinatorial optimization problem (i.e., the N-queen problem) using this network. We demonstrate the advantages of hysteresis by showing increased frequency of convergence to a solution, when the parameters associated with the activation function are varied.     

Knowledge‐Based Expert Control of Thalamic Neuron Firing Mode

This work sheds light on the possibility of using control strategies to set the parameters of electric stimulation, a commonly used technique in severe human central nervous system diseases. Currently, parameters of electric stimulation are set through a trial and error process, with a lot of undesirable side effects. Accordingly, and based on the problem of having a population of sick neurons embedded in a population of normal neurons, this work explores the possibility of using a control system based on the behavior of healthy neurons to set current parameters able to modify the electric behavior of sick neurons. Specifically, we posit a knowledge‐based expert control system that modifies the firing mode of a thalamic neuron by applying a control stimulation current, with the aim of making it fire in the same mode as a reference thalamic neuron. The controller parameters are tuned based on some characteristics of neurons that have to be determined through experiments before their application, but this controller does not require a detailed mathematical model of each neuron. Simulation results indicate that the proposed system satisfies the control objectives.     

深度单峰梯形神经网络

近年来，深度学习在许多领域取得了巨大的成功，其中，激活函数是深度学习取得巨大成功的关键因素之一。根据生物神经元特性，针对ReLU右侧响应无界问题，提出了单峰梯形线性单元（Single-Peaked Trapezoid Linear Unit，SPTLU）。SPTLU更加符合生物神经元特性，且取得了等同和超越ReLU的优异性能，实验表明在不同数据集上都取得了很好的效果，例如，数据集MNIST，Fashion-MNIST，SVHN，CALTECH101和CIFAR10。     

Evolutionary-modified fuzzy nearest-neighbor rule for pattern classification

This paper presents an improved version of the well-established k nearest neighbor (k-NN) and fuzzy NN (FNN), termed the multi-objective genetic-algorithm-modified FNN (MOGA-MFNN). The MFNN design problem is converted into a multi-modal objective maximization problem constrained by four objective functions. Thereafter, the associated parameter set of the MFNN and the feature attributes can be determined optimally and automatically via the non-dominated sorting genetic algorithm II. We introduce two new objective functions termed the Margin-I and Margin-II, which are used to improve the generalization capability of the MFNN for the unknown data, along with two existing performance functions: the geometric mean and the area under the receiver-operated characteristic curve for the training accuracy. Moreover, we proposed a novel data-dependent weight-assignment technique for local class membership functions of the MFNN. The technique enables the MFNN to determine its local neighbors adaptively through the MOGA algorithm. To expedite the classification, the MOGA-MFNN is implemented on a graphical processing unit (GPU), which significantly increases the computation speed. Furthermore, the local class-membership function of the MFNN can be computed in advance, rather than delaying it to the classification stage. This again can improve the classification speed. The MOGA-MFNN is evaluated on 20 datasets obtained from the repository of the University of California, Irvine (UCI). The experiments with rigorous statistical significance tests demonstrate that the proposed method performs competitively with the existing methods.     

Shannon小波混沌神经网络及其TSP(城市旅行商)问题的求解

混沌神经网络已经被证明是解决组合优化问题的有效工具.针对混沌神经网络的单调的激励函数,通过引入Shannon小波和Sigmoid函数加和组成的非单调激励函数,提出了一种新型的暂态混沌神经元模型.给出了该混沌神经元的倒分岔图和最大Lyapunov指数时间演化图,分析了其动力学特性.基于该模型,构造了一种暂态混沌神经网络,并将其应用于函数优化和组合优化问题.通过经典的10城市TSP验证了该暂态混沌神经网络的有效性.