ART神经网络的发展与应用

作为解决神经网络学习中“稳定性/可塑性两难问题“的一种尝试,ART神经网络一直备受关注.从最初的仅仅用于处理二值输入的非监督学习网络ART1,到具有有监督学习能力的ARTMAP网络,具有一定模糊逻辑运算能力的Fuzzy ART网络,再到现在对于ART网络中的各种尝试,ART神经网络不断发展、改进,以便适应不同的应用场合.本文着重介绍了ART网络的基本体系结构与发展历程,对于其应用领域加以概述.     

基于ART2神经网络算法改进的研究

ART2神经网络是按照自适应谐振理论建立的一种自组织、无监督的人工神经网络.通过分析经典自适应谐振神经网络聚类过程,针对传统ART2神经网络模型对分类的不确定性和网络权值模式漂移等不足,提出了基于算法改进的ART2神经网络模型.最后对改进的ART2神经网络进行了仿真,并与经典神经网络所做仿真的结果比较,验证了改进的ART2神经网络结构大大提高了分类的正确率,有效改善了模式漂移现象.降低了空间存储消耗.     

多神经网络在高维数据分类中的应用研究

针对BP神经网络在高维数据分类中存在训练时间长的缺点,提出一种新的多神经网络分类模型,该模型采用自组织特征映射(SOFM)网络对训练样本集进行无监督聚类,通过优化竞争层神经元权值,并以此训练BP神经网络实现数据分类.最后对自由手写数字样本进行识别,仿真实验表明,这一模型具有较强的分类能力和泛化能力.     

一种基于强化学习的ART2神经网络——RL-ART2

提出一种基于强化学习的ART2神经网络(RL-ART2),使其利用强化学习的特性通过与环境交互而无需训练样本即可进行在线学习,同时给出该神经网络的学习算法.当ART2神经网络运行时,通过内部竞争学习得到输出的分类模式,随后通过与环境交互得到神经网络分类模式的运行效果并对其进行评价.通过这种不断与环境的交互学习,当经过在线学习足够的时间和次数后,ART2神经网络即具有相当的识别率.移动机器人路径规划仿真实验表明,使用RL-ART2后与未使用前相比大大减少了机器人与障碍物的碰撞次数,实践证明该方法的合理性和有效性.     

一种基于椭圆区域的进化式模糊分类系统

介绍了一种进化式模糊分类系统.首先,介绍系统的基本特征及结构框架.然后,介绍了一种动态聚类算法,并运用动态聚类算法对输入的训练模式进行动态聚类,每一簇创建一条模糊规则.规则所对应的区域为类椭圆形区域.规则调整的策略是连续改变模糊分类规则的一个参数,使得分类系统对训练模式识别率不能再提高,对不能达到要求的调整,采用遗传算法进行调整.分析了规则调整的方法,给出了调整算法,也介绍了规则的插入和聚合策略.用两个典型的数据集来评测研究的系统,研究的分类系统在识别率与多层神经网络分类器相当,但训练时间远少于多层神经网络分类器的训练时间.     

模糊ART神经网络的识别算法及其应用

提出并设计了模糊ART神经网络的结构、学习规则和识别算法.为了把该算法应用于人脸识别,定义了相似函数和匹配搜索方法,通过向量柱状图提取人脸特征,并用模糊ART神经网络对向量柱状图生成的特征向量进行识别.仿真实验结果表明,对于快速学习和非快速学习,不同的人具有不同的识别率,各有不同的警戒参数值可以使神经网络到达在线最大识别率82.25%和86%.     

基于模糊聚类算法的神经网络集成

基于模糊聚类思想,提出了一种神经网络集成方法。利用隶属度函数,构造了一个分布函数,根据分布函数对训练数据进行抽样,用所抽得的数据作为个体神经网络的训练样本,多个个体神经网络构成神经网络集成,集成的输出采用相对多数投票法。理论分析和实验结果表明,该方法对模式分类能取得较好的效果。     

基于无监督学习算法训练径向基概率神经网络

本文研究了径向基概率神经网络(Radial Basis Probabilistic Neural Networks,RBPNN)的一种新的无监督学习算法,该算法整合了径向基概率神经网络的结构原理与动态聚类算法的特点,使得在对训练样本的聚类分析并正确划分其类别属性的同时,自动完成径向基概率神经网络的训练过程.本算法在对IRIS和双螺旋分类问题的应用中,取得了较好的分类效果,而且在推广能力方面,由本文算法训练的RBPNN要明显好于有监督训练的径向基函数神经网络(RBFNN).     

基于神经网络和模糊匹配算法的手写汉字预分类研究

对手写汉字识别问题,提出了一种在识别之前对手写汉字预分类的新方法,该方法用Neocognitron网提取字符笔画特征,然后采用有监督的扩展ART神经网络（SEART）产生一定数量的预分类组并通过基于模糊相似测量的匹配算法进行预分类。实验表明,该方法用于手写汉字分类效果良好,预分类正确率达到98.22％。     

基于模糊ART神经网络的在线人脸识别模型的设计和实现

本文描述了模糊ART神经网络的结构和特性,定义了相似函数和匹配搜索方法,通过去噪、去最小亮度和设计编码簿的方法产生人脸的特征向量图,以提取人脸特征,并用模糊ART神经网络对特征向量图进行识别.仿真实验证明,当选择合适的模糊ART神经网络参数后,该模型的在线最大识别率可以达到81.25%,离线识别率几乎为100%.     

基于统计分析的分阶段进化神经网络方法

基于统计分析和分阶段进化，提出一种新的进化神经网络设计方法．本文方法 的进化过程分三个阶段：第一阶段，首先按训练样本统计特性设计较小规模的神经网络；第 二阶段，引入所有训练样本，在第一阶段的基础上，逐步扩展网络结构，新添加的神经元总 是单独训练并以抵消原网络的输出误差为其训练目标，直至训练网络达到误差要求．第三阶 段，利用统计方法，将网络中非线性变换作用相似的神经元合并，简化网络结构．本文方法 一方面减轻了进化算法的压力,另一方面指出了网络进化的方向使得进化网络的学习过程不 再是黑箱问题．计算机仿真实验表明，该方法是有效的．     

Falcon: neural fuzzy control and decision systems using FKP and PFKP clustering algorithms.

Neural fuzzy networks proposed in the literature can be broadly classified into two groups. The first group is essentially fuzzy systems with self-tuning capabilities and requires an initial rule base to be specified prior to training. The second group of neural fuzzy networks, on the other hand, is able to automatically formulate the fuzzy rules from the numerical training data. Examples are the Falcon-ART, and the POPFNN family of networks. A cluster analysis is first performed on the training data and the fuzzy rules are subsequently derived through the proper connections of these computed clusters. This correspondence proposes two new networks: Falcon-FKP and Falcon-PFKP. They are extensions of the Falcon-ART network, and aimed to overcome the shortcomings faced by the Falcon-ART network itself, i.e., poor classification ability when the classes of input data are very similar to each other, termination of training cycle depends heavily on a preset error parameter, the fuzzy rule base of the Falcon-ART network may not be consistent Nauck, there is no control over the number of fuzzy rules generated, and learning efficiency may deteriorate by using complementarily coded training data. These deficiencies are essentially inherent to the fuzzy ART, clustering technique employed by the Falcon-ART network. Hence, two clustering techniques--Fuzzy Kohonen Partitioning (FKP) and its pseudo variant PFKP, are synthesized with the basic Falcon structure to compute the fuzzy sets and to automatically derive the fuzzy rules from the training data. The resultant neural fuzzy networks are Falcon-FKP and Falcon-PFKP, respectively. These two proposed networks have a lean and efficient training algorithm and consistent fuzzy rule bases. Extensive simulations are conducted using the two networks and their performances are encouraging when benchmarked against other neural and neural fuzzy systems.     

Analyzing the performance of fuzzy cognitive maps with non-linear hebbian learning algorithm in predicting autistic disorder

The soft computing technique of fuzzy cognitive maps (FCM) for modeling and predicting autistic spectrum disorder has been proposed. The FCM models the behavior of a complex system and is used to develop new knowledge based system applications. FCM combines the robust properties of fuzzy logic and neural networks. To overwhelm the limitations and to improve the efficiency of FCM, a good learning method of unsupervised training could be applied. A decision system based on human knowledge and experience with a FCM trained using unsupervised non-linear hebbian learning algorithm is proposed here. Through this work the hebbian algorithm on non-linear units is used for training FCMs for the autistic disorder prediction problem. The investigated approach serves as a guide in determining the prognosis and in planning the appropriate therapies to special children.     

A fuzzy neighborhood-based training algorithm for feedforward neural networks

In this work we present a new hybrid algorithm for feedforward neural networks, which combines unsupervised and supervised learning. In this approach, we use a Kohonen algorithm with a fuzzy neighborhood for training the weights of the hidden layers and gradient descent method for training the weights of the output layer. The goal of this method is to assist the existing variable learning rate algorithms. Simulation results show the effectiveness of the proposed algorithm compared with other well-known learning methods.     

Adaptive resonance neural classifier for identification of gases/odours using an integrated sensor array

A new approach to intelligent gas sensor (IGS) design using a classifier based on adaptive resonance theory (ART) artificial neural network (ANN) is presented. Using published data of sensor arrays fabricated and characterised at our laboratory, we demonstrate excellent gas/odour identification performance of our classifier for a 4-gas, 4-sensor system to identify individual gas/odour. Since the ART neural network is a self-organising classifier trained in the unsupervised mode, it avoids the drawbacks associated with static feedforward neural networks trained with a locally optimal backpropagation-type training algorithms applied by researchers in the recent past. The ART neural network offers easy implementability and real time performance in addition to giving excellent classification accuracy as demonstrated by our experiments.     

Parallel implementation of the fuzzy ARTMAP neural network paradigm on a hypercube

Abstract: The recent surge of interest in connectionist models arose through the availability of high speed parallel supercomputers and the advent of new learning algorithms. The computations performed on concurrent architectures are less costly than similar ones performed on sequential machines. In this paper, the design and implementation of a parallel version of fuzzy ARTMAP (Carpenter et al. 1992), which encompasses both neural and fuzzy logic, is discussed. Fuzzy ARTMAP is a supervised learning algorithm utilising two fuzzy ART modules and an associated mapping network. A simplified version of fuzzy ARTMAP (SFAM) was designed by incorporating a simplification of the match tracking concept on unsupervised fuzzy ART paradigms. The proposed simplified version consists of only one fuzzy ART module and an associated mapping network. A parallel fuzzy ARTMAP (PFAM) algorithm is then designed and implemented on a hypercube simulator (iPSC). The algorithm is parallelised for any architecture and, with the exception of issues related to communications, the implementation remains the same on any type of parallel machine. PFAM enjoys the advantage of reduced training time that makes the algorithm a successful candidate for applications that require both online testing and training. Such applications can range from underwater sonar detection and chemical plant processing control to nuclear reactor process control, flexible manufacturing and systems analysis.     

Growing fuzzy topology adaptive resonance theory models with a push-pull learning algorithm

A new incrementally growing neural network model, called the growing fuzzy topology ART (GFTART) model, is proposed based on integrating the conventional fuzzy ART model with the incremental topology-preserving mechanism of the growing cell structure (GCS) model. This is in addition, to a new training algorithm, called the push-pull learning algorithm. The proposed GFTART model has two purposes: First, to reduce the proliferation of incrementally generated nodes in the F2 layer by the conventional fuzzy ART model based on replacing each F2 node with a GCS. Second, to enhance the class-dependent clustering representation ability of the GCS model by including the categorization property of the conventional fuzzy ART model. In addition, the proposed push-pull training algorithm enhances the cluster discriminating property and partially improves the forgetting problem of the training algorithm in the GCS model.     

Evaluation of unsupervised feature extraction neural networks for face recognition

In this paper, new appearances based on neural networks (NN) algorithms are presented for face recognition. Face recognition is subdivided into two main stages: feature extraction and classifier. The suggested NN algorithms are the unsupervised Sanger principal component neural network (Sanger PCNN) and the self-organizing feature map (SOFM), which will be applied for features extraction of the frontal view of a face image. It is of interest to compare the unsupervised network with the traditional Eigenfaces technique. This paper presents an experimental comparison of the statistical Eigenfaces method for feature extraction and the unsupervised neural networks in order to evaluate the classification accuracies as comparison criteria. The classifier is done by the multilayer perceptron (MLP) neural network. Overcoming of the problem of the finite number of training samples per person is discussed. Experimental results are implemented on the Olivetti Research Laboratory database that contains variability in expression, pose, and facial details. The results show that the proposed method SOFM/MLP neural network is more efficient and robust than the Sanger PCNN/MLP and the Eigenfaces/MLP, when used a few number of training samples per person. As a result, it would be more applicable to utilize the SOFM/MLP NN in order to accomplish a higher level of accuracy within a recognition system.     

Integration of supervised ART-based neural networks with a hybrid genetic algorithm

In this paper, two evolutionary artificial neural network (EANN) models that are based on integration of two supervised adaptive resonance theory (ART)-based artificial neural networks with a hybrid genetic algorithm (HGA) are proposed. The search process of the proposed EANN models is guided by a knowledge base established by ART with respect to the training data samples. The EANN models explore the search space for “coarse” solutions, and such solutions are then refined using the local search process of the HGA. The performances of the proposed EANN models are evaluated and compared with those from other classifiers using more than ten benchmark data sets. The applicability of the EANN models to a real medical classification task is also demonstrated. The results from the experimental studies demonstrate the effectiveness and usefulness of the proposed EANN models in undertaking pattern classification problems.     

Constructive feedforward ART clustering networks. II

For pt.I see ibid., p.645-61 (2002). Part I of this paper defines the class of constructive unsupervised on-line learning simplified adaptive resonance theory (SART) clustering networks. Proposed instances of class SART are the symmetric fuzzy ART (S-Fuzzy ART) and the Gaussian ART (GART) network. In Part II of our work, a third network belonging to class SART, termed fully self-organizing SART (FOSART), is presented and discussed. FOSART is a constructive, soft-to-hard competitive, topology-preserving, minimum-distance-to-means clustering algorithm capable of: 1) generating processing units and lateral connections on an example-driven basis and 2) removing processing units and lateral connections on a minibatch basis. FOSART is compared with Fuzzy ART, S-Fuzzy ART, GART and other well-known clustering techniques (e.g., neural gas and self-organizing map) in several unsupervised learning tasks, such as vector quantization, perceptual grouping and 3-D surface reconstruction. These experiments prove that when compared with other unsupervised learning networks, FOSART provides an interesting balance between easy user interaction, performance accuracy, efficiency, robustness, and flexibility