采用资格迹的神经网络学习控制算法

强化学习是解决自适应问题的重要方法,被广泛地应用于连续状态下的学习控制,然而存在效率不高和收敛速度较慢的问题.在运用反向传播(back propagation,BP)神经网络基础上,结合资格迹方法提出一种算法,实现了强化学习过程的多步更新.解决了输出层的局部梯度向隐层节点的反向传播问题,从而实现了神经网络隐层权值的快速更新,并提供一个算法描述.提出了一种改进的残差法,在神经网络的训练过程中将各层权值进行线性优化加权,既获得了梯度下降法的学习速度又获得了残差梯度法的收敛性能,将其应用于神经网络隐层的权值更新,改善了值函数的收敛性能.通过一个倒立摆平衡系统仿真实验,对算法进行了验证和分析.结果显示,经过较短时间的学习,本方法能成功地控制倒立摆,显著提高了学习效率.     

Adaptive transfer learning in deep neural networks: Wind power prediction using knowledge transfer from region to region and between different task domains

Transfer learning (TL) in deep neural networks is gaining importance because, in most of the applications, the labeling of data is costly and time consuming. Additionally, TL also provides an effective weight initialization strategy for deep neural networks. This paper introduces the idea of adaptive TL in deep neural networks (ATL‐DNN) for wind power prediction. Specifically, we show in case of wind power prediction that adaptive TL of the deep neural networks system can be adaptively modified as regards training on a different wind farm is concerned. The proposed ATL‐DNN technique is tested for short‐term wind power prediction, where continuously arriving information has to be exploited. Adaptive TL not only helps in providing good weight initialization, but also in utilizing the incoming data for effective learning. Additionally, the proposed ATL‐DNN technique is shown to transfer knowledge between different task domains (wind power to wind speed prediction) and from one region to another region. The simulation results show that the proposed ATL‐DNN technique achieves average values of 0.0637, 0.0986, and 0.0984 for the mean absolute error, root mean squared error, and standard deviation error, respectively.     

非光滑凸情形Adam型算法的最优个体收敛速率

Adam是目前深度神经网络训练中广泛采用的一种优化算法框架,同时使用了自适应步长和动量技巧,克服了SGD的一些固有缺陷。但即使对于凸优化问题,目前Adam也只是在线学习框架下给出了和梯度下降法一样的regret界,动量的加速特性并没有得到体现。这里针对非光滑凸优化问题,通过巧妙选取动量和步长参数,证明了Adam的改进型具有最优的个体收敛速率,从而说明了Adam同时具有自适应和加速的优点。通过求解 ${l_1}$ 范数约束下的hinge损失问题,实验验证了理论分析的正确性和在算法保持稀疏性方面的良好性能。     

基于改进YOLOv3网络的学生特定行为识别

为了提高卷积神经网络在学生行为识别应用的检测精度,本文使用K-means聚类对特有数据集进行聚类得到更适应的anchor box,并且提出一种基于改进损失函数的YOLOv3网络模型。该网络模型将原有的平方和损失函数权重进行动态转化,注重计算连续变量的损失。新的损失函数能有效减低Sigmoid函数梯度消失的影响,使模型收敛更加快速。实验结果表明,基于改进损失函数的深度卷积神经网络应用对“抬头”“低头”“说话”3种姿态的识别均有提高。     

Multi-Scale Deep Cascade Bi-Forest for Electrocardiogram Biometric Recognition

Electrocardiogram (ECG) biometric recognition has emerged as a hot research topic in the past decade.Although some promising results have been reported,especially using sparse representation learning (SRL) and deep neural network,robust identification for small-scale data is still a challenge.To address this issue,we integrate SRL into a deep cascade model,and propose a multi-scale deep cascade bi-forest (MDCBF) model for ECG biometric recognition.We design the bi-forest based feature generator by fusing L1-norm sparsity and L2-norm collaborative representation to efficiently deal with noise.Then we propose a deep cascade framework,which includes multi-scale signal coding and deep cascade coding.In the former,we design an adaptive weighted pooling operation,which can fully explore the discriminative information of segments with low noise.In deep cascade coding,we propose level-wise class coding without backpropagation to mine more discriminative features.Extensive experiments are conducted on four small-scale ECG databases,and the results demonstrate that the proposed method performs competitively with state-of-the-art methods.     

基于AdaGrad的自适应NAG方法及其最优个体收敛性

与梯度下降法相比,自适应梯度下降方法(AdaGrad)利用过往平方梯度的算数平均保存了历史数据的几何信息,在处理稀疏数据时获得了更紧的收敛界.另一方面,Nesterov加速梯度方法(Nesterov's accelerated gradient,NAG)在梯度下降法的基础上添加了动量运算,在求解光滑凸优化问题时具有数量...     

AdaSVRG：自适应学习率加速SVRG

在深度学习任务中,随机方差衰减梯度法通过降低随机梯度方差,因此,其具有较好的稳定性和较高的计算效率。然而,这类方法在学习过程中均使用恒定的学习率,降低了随机方差衰减梯度法的计算效率。基于随机方差衰减梯度法,借鉴动量加速思想并对梯度估计采取加权平均策略,对学习率利用历史梯度信息进行自动调整,提出了自适应随机方差衰减梯度法。基于MNIST和CIFAR-10数据集,验证提出的自适应随机方差衰减梯度法的有效性。实验结果表明,自适应随机方差衰减梯度法在收敛速度和稳定性方面优于随机方差衰减梯度法和随机梯度下降法。     

卷积神经网络的聚焦均方损失函数设计

为了提高卷积神经网络在人体姿势估计任务上的精度,提出了一种基于均方损失函数（Mean Squared Error,MSE）的改进损失函数来处理网络学习中回归热点图的前景（高斯核）和背景之间像素点不均衡问题,根据前景与背景不同像素点值对损失函数赋予不同权重,并将其命名为聚焦均方损失函数（Focus Mean Squared Error,FMSE）.与均方损失函数相比,我们提出的聚焦均方损失函数可以有效地减少前景和背景之间像素点不均衡对网络性能的影响,帮助网络定位关键点的空间位置,提升了网络性能,并使得训练阶段中损失函数收敛速度更快.并在公开数据集上进行实验,以验证我们所提出的聚焦均方损失函数的有效性.     

基于梯度估计的非线性系统最优控制及仿真

基于自适应动态规划（ADP）执行-评价结构,应用神经网络（NN）对非线性系统进行最优控制求解.首先提出所求解非线性系统的一般形式;其次给定二次正定性能指标,求其哈密尔顿函（HJB）函数;分别应用神经网络对执行-评价结构中的性能指标和最优控制进行逼近,神经网络权重参数应用梯度法求得,从而可以求得其最有控制策略.而且对执行机构和评价机构神经网络权重参数的收敛性以及系统总体的稳定性进行了详细的分析,证明所求控制策略可以使系统稳定;最后,用仿真结果来验证所提出的方法的可行性.     

针对卷积神经网络流量分类器的对抗样本攻击防御

随着深度学习的兴起,深度神经网络被成功应用于多种领域,但研究表明深度神经网络容易遭到对抗样本的恶意攻击.作为深度神经网络之一的卷积神经网络(CNN)目前也被成功应用于网络流量的分类问题,因此同样会遭遇对抗样本的攻击.为提高CNN网络流量分类器防御对抗样本的攻击,本文首先提出批次对抗训练方法,利用训练过程反向传播误差的特...     

基于神经元网和带死区的最小二乘算法的非线性离散时间系统的自适应控制

针对非线性离散时间系统,提出了一种用带死区的最小二乘算法去调节神经网参数的算法,同其他算法相比,这种算法具有非常高的收敛速度.对于这种自适应控制算法,证明了闭环系统的所有信号是有界的,跟踪误差收敛到以零为原点的球中.     

加权多尺度基本形式及应用

为了使多值图像的局部变化信息适合单值图像的处理,提出了加权多尺度基本形式,其中单值图像被赋予不同的白适应权重,单值图像的变化量越大,对应的权重也就越高．将该方法分别应用于图像融合、彩色图像增强、各向异性扩散和彩色图像复原4个方面．与基于多尺度基本形式的方法及其他方法相比较,文中方法取得了更好的应用效果．     

改进的弹性网模型在深度神经网络中的应用

由于具有较高的模型复杂度，深层神经网络容易产生过拟合问题，为了减少该问题对网络性能的不利影响，提出一种基于改进的弹性网模型的深度学习优化方法。首先，考虑到变量之间的相关性，对弹性网模型中的L1范数的不同变量进行自适应加权，从而得到L2范数与自适应加权的L1范数的线性组合。其次，将改进的弹性网络模型与深度学习的优化模型相结合，给出在这种新正则项约束下求解神经网络参数的过程。然后，推导出改进的弹性网模型在神经网络优化中具有群组选择能力和Oracle性质，进而从理论上保证该模型是一种更加鲁棒的正则化方法。最后，在多个回归问题和分类问题的实验中，相对于L1、L2和弹性网正则项，该方法的回归测试误差可分别平均降低87.09、88.54和47.02，分类测试准确度可分别平均提高3.98、2.92和3.58个百分点。由此，在理论和实验两方面验证了改进的弹性网模型可以有效地增强深层神经网络的泛化能力，提升优化算法的性能，解决深度学习的过拟合问题。     

BAS-ADAM: An ADAM Based Approach to Improve the Performance of Beetle Antennae Search Optimizer

In this paper, we propose enhancements to Beetle Antennae search (BAS) algorithm, called BAS-ADAM, to smoothen the convergence behavior and avoid trapping in local-minima for a highly non-convex objective function. We achieve this by adaptively adjusting the step-size in each iteration using the adaptive moment estimation (ADAM) update rule. The proposed algorithm also increases the convergence rate in a narrow valley. A key feature of the ADAM update rule is the ability to adjust the step-size for each dimension separately instead of using the same step-size. Since ADAM is traditionally used with gradient-based optimization algorithms, therefore we first propose a gradient estimation model without the need to differentiate the objective function. Resultantly, it demonstrates excellent performance and fast convergence rate in searching for the optimum of non-convex functions. The efficiency of the proposed algorithm was tested on three different benchmark problems, including the training of a high-dimensional neural network. The performance is compared with particle swarm optimizer (PSO) and the original BAS algorithm.      

Training pi-sigma network by online gradient algorithm with penalty for small weight update

A pi-sigma network is a class of feedforward neural networks with product units in the output layer. An online gradient algorithm is the simplest and most often used training method for feedforward neural networks. But there arises a problem when the online gradient algorithm is used for pi-sigma networks in that the update increment of the weights may become very small, especially early in training, resulting in a very slow convergence. To overcome this difficulty, we introduce an adaptive penalty term into the error function, so as to increase the magnitude of the update increment of the weights when it is too small. This strategy brings about faster convergence as shown by the numerical experiments carried out in this letter.     

粒子群优化算法在神经网络控制中的应用

考虑粒子群优化算法在不确定系统的自适应控制中的应用。神经网络在不确定系统的自适应控制中起着重要作用。但传统的梯度下降法训练神经网络时收敛速度慢,容易陷入局部极小,且对网络的初始权值等参数极为敏感。为了克服这些缺点,提出了一种基于粒子群算法优化的RBF神经网络整定PID的控制策略。首先,根据粒子群算法的基本原理提出了优化得到RBF神经网络输出权、节点中心和节点基宽参数的初值的算法。其次,再利用梯度下降法对控制器参数进一步调节。将传统的神经网络控制与基于粒子群优化的神经网络控制进行了对比,结果表明,后者有更好逼近精度。以PID控制器参数整定为例,对一类非线性控制系统进行了仿真。仿真结果表明基于粒子群优化的神经网络控制具有较强的鲁棒性和自适应能力。     

Learning Algorithm and Retrieval Process for the Multiple Classes Random Neural Network Model

Gelenbe has modeled neural networks using an analogy with queuing theory. This model (called Random Neural Network) calculates the probability of activation of the neurons in the network. Recently, Fourneau and Gelenbe have proposed an extension of this model, called multiple classes random neural network model. The purpose of this paper is to describe the use of the multiple classes random neural network model to learn patterns having different colors. We propose a learning algorithm for the recognition of color patterns based upon non-linear equations of the multiple classes random neural network model using gradient descent of a quadratic error function. In addition, we propose a progressive retrieval process with adaptive threshold values. The experimental evaluation shows that the learning algorithm provides good results.     

On the computational power of winner-take-all

This article initiates a rigorous theoretical analysis of the computational power of circuits that employ modules for computing winner-take-all. Computational models that involve competitive stages have so far been neglected in computational complexity theory, although they are widely used in computational brain models, artificial neural networks, and analog VLSI. Our theoretical analysis shows that winner-take-all is a surprisingly powerful computational module in comparison with threshold gates (also referred to as McCulloch-Pitts neurons) and sigmoidal gates. We prove an optimal quadratic lower bound for computing winner-take-all in any feedforward circuit consisting of threshold gates. In addition we show that arbitrary continuous functions can be approximated by circuits employing a single soft winner-take-all gate as their only nonlinear operation. Our theoretical analysis also provides answers to two basic questions raised by neurophysiologists in view of the well-known asymmetry between excitatory and inhibitory connections in cortical circuits: how much computational power of neural networks is lost if only positive weights are employed in weighted sums and how much adaptive capability is lost if only the positive weights are subject to plasticity.     

图像分类的深度卷积神经网络模型综述

图像分类是计算机视觉中的一项重要任务,传统的图像分类方法具有一定的局限性。随着人工智能技术的发展,深度学习技术越来越成熟,利用深度卷积神经网络对图像进行分类成为研究热点,图像分类的深度卷积神经网络结构越来越多样,其性能远远好于传统的图像分类方法。本文立足于图像分类的深度卷积神经网络模型结构,根据模型发展和模型优化的历程,将深度卷积神经网络分为经典深度卷积神经网络模型、注意力机制深度卷积神经网络模型、轻量级深度卷积神经网络模型和神经网络架构搜索模型等4类,并对各类深度卷积神经网络模型结构的构造方法和特点进行了全面综述,对各类分类模型的性能进行了对比与分析。虽然深度卷积神经网络模型的结构设计越来越精妙,模型优化的方法越来越强大,图像分类准确率在不断刷新的同时,模型的参数量也在逐渐降低,训练和推理速度不断加快。然而深度卷积神经网络模型仍有一定的局限性,本文给出了存在的问题和未来可能的研究方向,即深度卷积神经网络模型主要以有监督学习方式进行图像分类,受到数据集质量和规模的限制,无监督式学习和半监督学习方式的深度卷积神经网络模型将是未来的重点研究方向之一;深度卷积神经网络模型的速度和资源消耗仍不尽人意,应用于移动式设备具有一定的挑战性;模型的优化方法以及衡量模型优劣的度量方法有待深入研究;人工设计深度卷积神经网络结构耗时耗力,神经架构搜索方法将是未来深度卷积神经网络模型设计的发展方向。     

深度生成模型综述

通过学习可观测数据的概率密度而随机生成样本的生成模型在近年来受到人们的广泛关注,网络结构中包含多个隐藏层的深度生成式模型以更出色的生成能力成为研究热点,深度生成模型在计算机视觉、密度估计、自然语言和语音识别、半监督学习等领域得到成功应用,并给无监督学习提供了良好的范式.本文根据深度生成模型处理似然函数的不同方法将模型分...