基于样本密度和分类误差率的增量学习矢量量化算法研究

作为一种简单而成熟的分类方法, K最近邻(K nearest neighbor, KNN)算法在数据挖掘、模式识别等领域获得了广泛的应用, 但仍存在计算量大、高空间消耗、运行时间长等问题. 针对这些问题, 本文在增量学习型矢量量化(Incremental learning vector quantization, ILVQ)的单层竞争学习基础上, 融合样本密度和分类误差率的邻域思想, 提出了一种新的增量学习型矢量量化方法, 通过竞争学习策略对代表点邻域实现自适应增删、合并、分裂等操作, 快速获取原始数据集的原型集, 进而在保障分类精度基础上, 达到对大规模数据的高压缩效应. 此外, 对传统近邻分类算法进行了改进, 将原型近邻集的样本密度和分类误差率纳入到近邻判决准则中. 所提出算法通过单遍扫描学习训练集可快速生成有效的代表原型集, 具有较好的通用性. 实验结果表明, 该方法同其他算法相比较, 不仅可以保持甚至提高分类的准确性和压缩比, 且具有快速分类的优势.     

Local feature weighting in nearest prototype classification.

The distance metric is the corner stone of nearest neighbor (NN)-based methods, and therefore, of nearest prototype (NP) algorithms. That is because they classify depending on the similarity of the data. When the data is characterized by a set of features which may contribute to the classification task in different levels, feature weighting or selection is required, sometimes in a local sense. However, local weighting is typically restricted to NN approaches. In this paper, we introduce local feature weighting (LFW) in NP classification. LFW provides each prototype its own weight vector, opposite to typical global weighting methods found in the NP literature, where all the prototypes share the same one. Providing each prototype its own weight vector has a novel effect in the borders of the Voronoi regions generated: They become nonlinear. We have integrated LFW with a previously developed evolutionary nearest prototype classifier (ENPC). The experiments performed both in artificial and real data sets demonstrate that the resulting algorithm that we call LFW in nearest prototype classification (LFW-NPC) avoids overfitting on training data in domains where the features may have different contribution to the classification task in different areas of the feature space. This generalization capability is also reflected in automatically obtaining an accurate and reduced set of prototypes.     

A novel feature selection algorithm based on LVQ hypothesis margin

Feature selection has been widely discussed as an important preprocessing step in machine learning and data mining. In this paper, a new feature selection evaluation criterion based on low-loss learning vector quantization (LVQ) classification is proposed. Based on the evaluation criterion, a feature selection algorithm that optimizes the hypothesis margin of LVQ classification through minimizing its loss function is presented. Some experiments that are compared with well-known SVM-RFE and Relief are carried out on 4 UCI data sets using Naive Bayes and RBF Network classifier. Experimental results show that new algorithm achieves similar or even higher performance than Relief on all training data and has better or comparable performance than SVM-RFE.     

Maximum margin multiple-instance feature weighting

Feature weighting is of considerable importance in machine learning due to its effectiveness to highlight relevant components and suppress irrelevant ones. In this paper, we focus on the feature weighting problem in a specific machine learning area: multiple-instance learning, and propose maximum margin multiple-instance feature weighting (M3IFW) to seek large classification margins in the weighted feature space. The designed M3IFW algorithm can be applied to both standard binary-class multiple-instance learning and the corresponding multi-class learning, and we abbreviate them to B-M3IFW (binary-class M3IFW) and M-M3IFW (multi-class M3IFW), respectively. Both B-M3IFW and M-M3IFW contain three kinds of unknown variables, i.e., positive prototypes, classification margins, and weighting coefficients. We utilize the coordinate ascent algorithm to update the three kinds of unknown variables, respectively and iteratively, and then perform classifications in the weighted feature space. Experiments conducted on synthetic and real-world datasets empirically demonstrate the effectiveness of M3IFW in improving classification accuracies.     

Soft nearest prototype classification

We propose a new method for the construction of nearest prototype classifiers which is based on a Gaussian mixture ansatz and which can be interpreted as an annealed version of learning vector quantization (LVQ). The algorithm performs a gradient descent on a cost-function minimizing the classification error on the training set. We investigate the properties of the algorithm and assess its performance for several toy data sets and for an optical letter classification task. Results show 1) that annealing in the dispersion parameter of the Gaussian kernels improves classification accuracy; 2) that classification results are better than those obtained with standard learning vector quantization (LVQ 2.1, LVQ 3) for equal numbers of prototypes; and 3) that annealing of the width parameter improved the classification capability. Additionally, the principled approach provides an explanation of a number of features of the (heuristic) LVQ methods.     

Performance evaluation of pattern classifiers for handwritten character recognition

This paper describes a performance evaluation study in which some efficient classifiers are tested in handwritten digit recognition. The evaluated classifiers include a statistical classifier (modified quadratic discriminant function, MQDF), three neural classifiers, and an LVQ (learning vector quantization) classifier. They are efficient in that high accuracies can be achieved at moderate memory space and computation cost. The performance is measured in terms of classification accuracy, sensitivity to training sample size, ambiguity rejection, and outlier resistance. The outlier resistance of neural classifiers is enhanced by training with synthesized outlier data. The classifiers are tested on a large data set extracted from NIST SD19. As results, the test accuracies of the evaluated classifiers are comparable to or higher than those of the nearest neighbor (1-NN) rule and regularized discriminant analysis (RDA). It is shown that neural classifiers are more susceptible to small sample size than MQDF, although they yield higher accuracies on large sample size. As a neural classifier, the polynomial classifier (PC) gives the highest accuracy and performs best in ambiguity rejection. On the other hand, MQDF is superior in outlier rejection even though it is not trained with outlier data. The results indicate that pattern classifiers have complementary advantages and they should be appropriately combined to achieve higher performance. Received: July 18, 2001 / Accepted: September 28, 2001     

Handwritten digit recognition: benchmarking of state-of-the-art techniques

This paper presents the results of handwritten digit recognition on well-known image databases using state-of-the-art feature extraction and classification techniques. The tested databases are CENPARMI, CEDAR, and MNIST. On the test data set of each database, 80 recognition accuracies are given by combining eight classifiers with ten feature vectors. The features include chaincode feature, gradient feature, profile structure feature, and peripheral direction contributivity. The gradient feature is extracted from either binary image or gray-scale image. The classifiers include the k-nearest neighbor classifier, three neural classifiers, a learning vector quantization classifier, a discriminative learning quadratic discriminant function (DLQDF) classifier, and two support vector classifiers (SVCs). All the classifiers and feature vectors give high recognition accuracies. Relatively, the chaincode feature and the gradient feature show advantage over other features, and the profile structure feature shows efficiency as a complementary feature. The SVC with RBF kernel (SVC-rbf) gives the highest accuracy in most cases but is extremely expensive in storage and computation. Among the non-SV classifiers, the polynomial classifier and DLQDF give the highest accuracies. The results of non-SV classifiers are competitive to the best ones previously reported on the same databases.     

Prototype-based minimum classification error/generalized probabilistic descent training for various speech units

In previous work we reported high classification rates for learning vector quantization (LVQ) networks trained to classify phoneme tokens shifted in time. It has since been shown that the framework of minimum classification error (MCE) and generalized probabilistic descent (GPD) can treat LVQ as a special case of a general method for gradient descent on a rigorously defined classification loss measure that closely reflects the misclassification rate. This framework allows us to extend LVQ into a prototype-based minimum error classifier (PBMEC) appropriate for the classification of various speech units which the original LVQ was unable to treat. Speech categories are represented using a prototype-based multi-state architecture incorporating a dynamic time warping procedure. We present results for the difficult E-set task, as well as for isolated word recognition for a vocabulary of 5240 words, that reveal clear gains in performance as a result of using PBMEC. In addition, we discuss the issue of smoothing the loss function from the perspective of increasing classifier robustness.     

熵约束广义学习矢量量化神经网络和软竞争学习算法

结合广义学习矢量量化神经网络的思想和信息论中的极大熵原理,提出了一种熵约束 广义学习矢量量化神经网络,利用梯度下降法导出其学习算法,该算法是软竞争格式的一种推 广.由于亏损因子和尺度函数被定义为同一个模糊隶属度函数,它可以有效地克服广义学习矢 量量化网络的模糊算法存在的问题.文中还给出熵约束广义学习矢量量化网络及其软竞争学习 算法的许多重要性质,以此为依据,讨论拉格朗日乘子的选取规则.     

Minimum-risk training for semi-Markov conditional random fields with application to handwritten Chinese/Japanese text recognition

Semi-Markov conditional random fields (semi-CRFs) are usually trained with maximum a posteriori (MAP) criterion which adopts the 0/1 cost for measuring the loss of misclassification. In this paper, based on our previous work on handwritten Chinese/Japanese text recognition (HCTR) using semi-CRFs, we propose an alternative parameter learning method by minimizing the risk on the training set, which has unequal misclassification costs depending on the hypothesis and the ground-truth. Based on this framework, three non-uniform cost functions are compared with the conventional 0/1 cost, and training data selection is incorporated to reduce the computational complexity. In experiments of online handwriting recognition on databases CASIA-OLHWDB and TUAT Kondate, we compared the performances of the proposed method with several widely used learning criteria, including conditional log-likelihood (CLL), softmax-margin (SMM), minimum classification error (MCE), large-margin MCE (LM-MCE) and max-margin (MM). On the test set (online handwritten texts) of ICDAR 2011 Chinese handwriting recognition competition, the proposed method outperforms the best system in competition.     

Classifier ensemble construction with rotation forest to improve medical diagnosis performance of machine learning algorithms

Improving accuracies of machine learning algorithms is vital in designing high performance computer-aided diagnosis (CADx) systems. Researches have shown that a base classifier performance might be enhanced by ensemble classification strategies. In this study, we construct rotation forest (RF) ensemble classifiers of 30 machine learning algorithms to evaluate their classification performances using Parkinson's, diabetes and heart diseases from literature.While making experiments, first the feature dimension of three datasets is reduced using correlation based feature selection (CFS) algorithm. Second, classification performances of 30 machine learning algorithms are calculated for three datasets. Third, 30 classifier ensembles are constructed based on RF algorithm to assess performances of respective classifiers with the same disease data. All the experiments are carried out with leave-one-out validation strategy and the performances of the 60 algorithms are evaluated using three metrics; classification accuracy (ACC), kappa error (KE) and area under the receiver operating characteristic (ROC) curve (AUC).Base classifiers succeeded 72.15%, 77.52% and 84.43% average accuracies for diabetes, heart and Parkinson's datasets, respectively. As for RF classifier ensembles, they produced average accuracies of 74.47%, 80.49% and 87.13% for respective diseases.RF, a newly proposed classifier ensemble algorithm, might be used to improve accuracy of miscellaneous machine learning algorithms to design advanced CADx systems.     

基于KNN和RVM的分类方法——KNN RVM分类器

针对相关向量机(RVM)算法分类精度低、核参数选择困难等问题,文中提出临界滑动阈值的概念并以其为基础将RVM与K近邻(KNN)算法结合构建分类器——KNN-RVM分类器。从理论上提出并证明KNN-RVM分类过程等价于带软间隔约束的支持向量机的分类过程、KNN-RVM分类器等价于每类只选一个代表点的1-NN分类器、KNN-RVM分类效果优于RVM这3个结论。对这3个不同数据集进行实验证明临界滑动阈值的临界性与滑动性及KNN-RVM分类器的准确性、适应性及全局最优性,提高分类精度,减轻算法对核参数的依赖性,进而证明KNN-RVM分类器是一种有效的分类器。     

Nearest prototype classifier designs: An experimental study

We compare eleven methods for finding prototypes upon which to base the nearest prototype classifier. Four methods for prototype selection are discussed: Wilson+Hart (a condensation+error‐editing method), and three types of combinatorial search—random search, genetic algorithm, and tabu search. Seven methods for prototype extraction are discussed: unsupervised vector quantization, supervised learning vector quantization (with and without training counters), decision surface mapping, a fuzzy version of vector quantization, c‐means clustering, and bootstrap editing. These eleven methods can be usefully divided two other ways: by whether they employ pre‐ or postsupervision; and by whether the number of prototypes found is user‐defined or “automatic.” Generalization error rates of the 11 methods are estimated on two synthetic and two real data sets. Offering the usual disclaimer that these are just a limited set of experiments, we feel confident in asserting that presupervised, extraction methods offer a better chance for success to the casual user than postsupervised, selection schemes. Finally, our calculations do not suggest that methods which find the “best” number of prototypes “automatically” are superior to methods for which the user simply specifies the number of prototypes. © 2001 John Wiley & Sons, Inc.     

Improving the effectiveness of RBF classifier based on a hybrid cost function

This paper presents a new minimum classification error (MCE)–mean square error (MSE) hybrid cost function to enhance the classification ability and speed up the learning process of radial basis function (RBF)-based classifier. Contributed by the MCE function, the proposed cost function enables the RBF-based classifier to achieve an excellent classification performance compared with the conventional MSE function. In addition, certain learning difficulties experienced by the MCE algorithm can be solved in an efficient and simple way. The presented results show that the proposed method exhibits a substantially higher convergence rate compared with the MCE function.     

A combination of CSP-based method with soft margin SVM classifier and generalized RBF kernel for imagery-based brain computer interface applications

Several methods utilizing common spatial pattern (CSP) algorithm have been presented for improving the identification of imagery movement patterns for brain computer interface applications. The present study focuses on improving a CSP-based algorithm for detecting the motor imagery movement patterns. A discriminative filter bank of CSP method using a discriminative sensitive learning vector quantization (DFBCSP-DSLVQ) system is implemented. Four algorithms are then combined to form three methods for improving the efficiency of the DFBCSP-DSLVQ method, namely the kernel linear discriminant analysis (KLDA), the kernel principal component analysis (KPCA), the soft margin support vector machine (SSVM) classifier and the generalized radial bases functions (GRBF) kernel. The GRBF is used as a kernel for the KLDA, the KPCA feature selection algorithms and the SSVM classifier. In addition, three types of classifiers, namely K-nearest neighbor (K-NN), neural network (NN) and traditional support vector machine (SVM), are employed to evaluate the efficiency of the classifiers. Results show that the best algorithm is the combination of the DFBCSP-DSLVQ method using the SSVM classifier with GRBF kernel (SSVM-GRBF), in which the best average accuracy, attained are 92.70% and 83.21%, respectively. Results of the Repeated Measures ANOVA shows the statistically significant dominance of this method at p <?0.05. The presented algorithms are then compared with the base algorithm of this study i.e. the DFBCSP-DSLVQ with the SVM-RBF classifier. It is concluded that the algorithms, which are based on the SSVM-GRBF classifier and the KLDA with the SSVM-GRBF classifiers give sufficient accuracy and reliable results.

     

Differential evolution based nearest prototype classifier with optimized distance measures for the features in the data sets

In this paper a further generalization of differential evolution based data classification method is proposed, demonstrated and initially evaluated. The differential evolution classifier is a nearest prototype vector based classifier that applies a global optimization algorithm, differential evolution, for determining the optimal values for all free parameters of the classifier model during the training phase of the classifier. The earlier version of differential evolution classifier that applied individually optimized distance measure for each new data set to be classified is generalized here so, that instead of optimizing a single distance measure for the given data set, we take a further step by proposing an approach where distance measures are optimized individually for each feature of the data set to be classified. In particular, distance measures for each feature are selected optimally from a predefined pool of alternative distance measures. The optimal distance measures are determined by differential evolution algorithm, which is also determining the optimal values for all free parameters of the selected distance measures in parallel. After determining the optimal distance measures for each feature together with their optimal parameters, we combine all featurewisely determined distance measures to form a single total distance measure, that is to be applied for the final classification decisions. The actual classification process is still based on the nearest prototype vector principle; A sample belongs to the class represented by the nearest prototype vector when measured with the above referred optimized total distance measure. During the training process the differential evolution algorithm determines optimally the class vectors, selects optimal distance metrics for each data feature, and determines the optimal values for the free parameters of each selected distance measure. Based on experimental results with nine well known classification benchmark data sets, the proposed approach yield a statistically significant improvement to the classification accuracy of differential evolution classifier.     

Soft learning vector quantization

Learning vector quantization (LVQ) is a popular class of adaptive nearest prototype classifiers for multiclass classification, but learning algorithms from this family have so far been proposed on heuristic grounds. Here, we take a more principled approach and derive two variants of LVQ using a gaussian mixture ansatz. We propose an objective function based on a likelihood ratio and derive a learning rule using gradient descent. The new approach provides a way to extend the algorithms of the LVQ family to different distance measure and allows for the design of "soft" LVQ algorithms. Benchmark results show that the new methods lead to better classification performance than LVQ 2.1. An additional benefit of the new method is that model assumptions are made explicit, so that the method can be adapted more easily to different kinds of problems.     

Maximum likelihood topographic map formation

We introduce a new unsupervised learning algorithm for kernel-based topographic map formation of heteroscedastic gaussian mixtures that allows for a unified account of distortion error (vector quantization), log-likelihood, and Kullback-Leibler divergence.     

On the Generalization Ability of GRLVQ Networks

We derive a generalization bound for prototype-based classifiers with adaptive metric. The bound depends on the margin of the classifier and is independent of the dimensionality of the data. It holds for classifiers based on the Euclidean metric extended by adaptive relevance terms. In particular, the result holds for relevance learning vector quantization (RLVQ) [4] and generalized relevance learning vector quantization (GRLVQ) [19].     

A prototype classification method and its use in a hybrid solution for multiclass pattern recognition

In this paper, we propose a prototype classification method that employs a learning process to determine both the number and the location of prototypes. This learning process decides whether to stop adding prototypes according to a certain termination condition, and also adjusts the location of prototypes using either the K-means (KM) or the fuzzy c-means (FCM) clustering algorithms. When the prototype classification method is applied, the support vector machine (SVM) method can be used to post-process the top-rank candidates obtained during the prototype learning or matching process. We apply this hybrid solution to handwriting recognition and address the convergence behavior and runtime consumption of the prototype construction process, and discuss how to combine our prototype classifier with SVM classifiers to form an effective hybrid classifier.