关于周期神经网络逼近阶的研究(英文)

借助于有关Fourier级数的Riesz平均构造出了一类含有一个隐含层的周期神经网络与平移网络,与已有的讨论相比较,在获得相同的逼近阶的情况下,此类网络的隐层单元要求较少的神经元个数．

非周期神经网络及平移网络在L_w^p中的逼近

设s≥d≥1为整数, 1≤p≤+∞,借助于正交多元代数多项式系而构造了一类s维网络算子,并用于逼近Lpw[-1,1]s中的函数,给出了逼近的上界以及当此算子为平移网络算子及神经网络算子时的导数型估计.

基于多周期组件时空神经网络的路网通行速度预测

针对当前路网通行速度预测方法存在的中长周期预测准确性和稳定性不足、自适应路网拓扑空间关系建模能力有待进一步提升等问题,以多尺度卷积算子及门控循环单元为核心单元,提出一种面向路网通行速度预测任务的多周期组件时空神经网络模型。首先,根据路网交通感知数据的周期特性,将其规约为周、日和近期这3种不同粒度的时间-空间-特征三维矩阵,并输入至3个共享网络结构的周期组件。其次,在每部分组件中,利用多尺度卷积核捕获多因素非线性相关性与不同空间视野大小的路网节点空间相关性。然后,对每个路网节点的时序特征使用门控循环单元提取交通数据长时依赖关系,引入残差学习框架,提高网络训练效率并防止梯度弥散。最后,自适应加权融合通过预测卷积层的每部分周期组件预测结果生成预测时段内路网交通通行速度。为验证所提方法的有效性,基于两个公开的交通状态数据集进行实验分析,并选取当前主流的深度神经网络模型作为对比基线模型。结果表明,所提方法在可接受的执行时间内,在两个数据集上平均绝对误差、平均平方误差和平均绝对百分比误差分别为2.55、3.94和10.75%,1.57、3.52和3.44%,在预测准确性与中长时多步预测稳定性方面均优于其他基准方法。

具分布时滞的周期运动细胞神经网络周期解的存在性(英文)

运用Mawhin连续性定理研究具分布时滞的周期运动细胞神经网络周期解的存在性,假设行为函数位于一带型区域内,激活函数位于两线性函数所夹的区域内。

一类具分布时滞的周期运动细胞神经网络周期解的存在性(英文)

运用迭合度理论和一些分析技巧研究了一类具分布时滞的周期运动细胞神经网络周期解的存在性.给出了要求更弱的判定周期解的存在性条件,从而改进了前人结论中周期解存在的相关条件.

基于多变量LSTM神经网络的澳大利亚大火预测研究

长短周期记忆神经网络(LSTM)受益于能够捕获长期依赖关系的特点,在许多实际应用中展现了优异的性能。该文构建了LSTM多变量数据驱动的预测模型,通过多变量输入的方式预测澳大利亚森林大火。首先使用多变量LSTM预测模型对日最高温度进行预测,并与反向传播(BP)神经网络以及ARIMA预测模型的结果进行对比。研究表明:以相关变量为输入的BP神经网络无法考虑时序变化规律,预测误差最大;以温度单变量为输入的ARIMA根据时序变化做出相应预测,预测效果较好;多变量LSTM预测模型综合考虑了多种因素的相互影响,同时结合了时间序列依赖关系,预测效果最好。最后通过多变量LSTM预测模型对某节点是否着火进行了预测,预测结果与实际值契合较好。总体来说,多变量LSTM预测模型对澳大利亚大火的预测结果可信。

基于SAE和LSTM RNN的多模态生理信号融合和情感识别研究

为了提高情感识别的分类准确率,提出一种将栈式自编码神经网络(SAE)和长短周期记忆单元循环神经网络(LSTM RNN)融合的多模态融合特征情感识别方法。该方法通过SAE对不同模态的生理特征进行信息融合和压缩,随后用LSTM RNN对长时间周期的融合进行情感分类识别。通过将该方法用到开源数据集中进行验证,得到情感分类准确率达到0.792 6。实验结果表明,SAE对多模态生理特征进行了有效融合,LSTM RNN能够有效地对长时间周期中的关键特征进行识别。

Global exponential periodicity of a class of impulsive neural networks

By the Lyapunov function method, combined with the inequality techniques, some criteria are established to ensure the existence, uniqueness and global exponential stability of the periodic solution for a class of impulsive neural networks. The results obtained only require the activation functions to be globally Lipschitz continuous without assuming their boundedness, monotonicity or differentiability. The conditions are easy to check in practice and they can be applied to design globally exponentially periodic impulsive neural networks.

Backstepping sliding mode control with self recurrent wavelet neural network observer for a novel coaxial twelve-rotor UAV

The robust attitude control for a novel coaxial twelve-rotor UAV which has much greater payload capacity,higher drive capability and damage tolerance than a quad-rotor UAV is studied. Firstly,a dynamical and kinematical model for the coaxial twelve-rotor UAV is designed. Considering model uncertainties and external disturbances,a robust backstepping sliding mode control( BSMC) with self recurrent wavelet neural network( SRWNN) method is proposed as the attitude controller for the coaxial twelve-rotor. A combinative algorithm of backstepping control and sliding mode control has simplified design procedures with much stronger robustness benefiting from advantages of both controllers. SRWNN as the uncertainty observer is able to estimate the lumped uncertainties effectively.Then the uniformly ultimate stability of the twelve-rotor system is proved by Lyapunov stability theorem. Finally,the validity of the proposed robust control method adopted in the twelve-rotor UAV under model uncertainties and external disturbances are demonstrated via numerical simulations and twelve-rotor prototype experiments.

Convergence and Periodicity of Solutions for a Discrete Model

The discrete-time network model of two neurons with function f（u） ={1,u∈[0,σ] 0,U∈[0,σ]is considered. We obtain some sufficient conditions that every solution of system is convergent or periodic.

Image Segmentation Based on Period Difference of the Oscillation

A new method for image segmentation based on pulse neural network is proposed. Every neuron in the network represents one pixel in the image and the network is locally connected. Each group of the neurons that correspond to each object synchronizes while different groups of the neurons oscillate at different period. Applying this period difference, different objects are divided. In addition to simulation, an analysis of the mechanism of the method is presented in this paper.

Coherence resonance and bi-resonance by time-periodic coupling strength in Hodgkin-Huxley neuron networks

We study the effect of time-periodic coupling strength on the spiking coherence of Newman-Watts networks of Hodgkin-Huxley(HH) neurons with non-Gaussian noise.It is found that the spiking can exhibit coherence resonance(CR) when the extent of deviation of non-Gaussian noise from Gaussian noise and the amplitude of the coupling strength are varied.In particular,coherence bi-resonance(CBR) is observed when the frequency of the coupling strength is varied,and the CBR is always observed when the frequency is equal to,or a multiple of,the spiking period,manifesting as the locking between the frequencies of the spiking and the coupling strength.The results show that a time-periodic coupling strength may play a more constructive and efficient role in enhancing the spiking coherence of the neuronal networks than a constant coupling strength.These findings provide insight into the role of time-periodic coupling strength for enhancing the time precision of information processing in neuronal networks.

Emergence of target waves in neuronal networks due to diverse forcing currents

The electric activities of neurons could be changed when ion channel block occurs in the neurons.External forcing currents with diversity are imposed on the regular network of Hodgkin-Huxley(HH) neuron,and target waves are induced to occupy the network.The forcing current I1 is imposed on neurons in a local region with m 0 ×m 0 nodes in the network,neurons in other nodes are imposed with another forcing current I2.Target wave could be developed to occupy the network when the gradient forcing current(I1-I2) exceeds certain threshold,and the formation of target wave is independent of the selection of boundary condition.It is also found that the developed target wave can decrease the negative effect of ion channel block and suppress the spiral wave,and thus channel noise is also considered.The potential mechanism of formation of target wave could be that the gradient forcing current(I1-I2) generates quasi-periodical signal in local area,and the propagation of quasi-periodical signal induces target-like wave due to mutual coupling among neurons in the network.

When a smooth self-map of a semi-simple Lie group can realize the least number of periodic points

There are two algebraic lower bounds of the number of n-periodic points of a self-map f ： M →4 M of a compact smooth manifold of dimension at least 3： NFn（f） = min{#Fix（gn）;g - f; g continuous} and NJDn（f） = min{#Fix（gn）;g - f; g smooth}. In general, NJDn（f） may be much greater than NFn（f）. We show that for a self-map of a semi-simple Lie group, inducing the identity fundamental group homomorphism, the equality NFn（f） = NJDn（f） holds for all n →← all eigenvalues of a quotient cohomology homomorphism induced by f have moduli ≤ 1.

Exponential synchronization for neural networks with mixed time-varying delays via periodically intermittent control

This paper deals with exponential synchronization for a class of neural networks with mixed time-varying delays via periodically intermittent control. Some novel and effective exponential synchronization criteria are derived by constructing Lyapunov functional and applying some analysis techniques. These results presented in this paper generalize and improve many known results. Finally, this paper presents an illustrative example and uses the simulated results to show the feasibility and effectiveness of the proposed scheme.