基于标记特定特征和相关性的ML-KNN改进算法

目前大部分已经存在的多标记学习算法在模型训练过程中所采用的共同策略是基于相同的标记属性特征集合预测所有标记类别.但这种思路并未对每个标记所独有的标记特征进行考虑.在标记空间中,这种标记特定的属性特征对于区分其它类别标记和描述自身特性是非常有帮助的信息.针对这一问题,本文提出了基于标记特定特征和相关性的ML-KNN改进算法MLF-KNN.不同于之前的多标记算法直接在原始训练数据集上进行操作,而是首先对训练数据集进行预处理,为每一种标记类别构造其特征属性,在得到的标记属性空间上进一步构造L₁-范数并进行优化从而引入标记之间的相关性,最后使用改进后的ML-KNN算法进行预测分类.实验结果表明,在公开数据集image和yeast上,本文提出的算法MLF-KNN分类性能优于ML-KNN,同时与其它另外3种多标记学习算法相比也表现出一定的优越性.     

Analyzing the statistical features of CIXL2 crossover offspring

We cannot deny the effort that the scientific community is devoting to the explanation of the features of the crossover operator of real-coded genetic algorithms and its effect over the evolutive process. This paper is another step in that direction, we analyze the behavior of the Confidence Interval Based Crossover using L₂ Norm (CIXL2). This crossover is based on the learning of the statistical features of localization and dispersion of the best individuals of the population. The crossover obtains, by means of a L₂ norm, the estimators of the parameters of localization and dispersion of the distributions of the fittest individuals. From this estimation three virtual parents are created using the localization parameter and the lower and upper bounds of the bilateral confidence intervals of the gene values of the best individuals of the population. This paper studies the statistical features of the offspring generated by this crossover and corroborates this study showing the behavior of the crossover in a set of test functions.     

基于对比散度受限玻尔兹曼机深度学习的产品评论情感分析

针对目前大部分情感分析技术需要人工标注建立情感词典提取情感特征的问题,提出一种基于对比散度-受限玻尔兹曼机(CD-RBM)深度学习的产品评论情感分析方法。该方法在对产品评论时进行数据预处理并利用词袋模型产生产品评论的向量表示,然后通过CD-RBM提取产品评论的情感特征,最后结合支持向量机(SVM)将提取出来的情感特征进行文本情感分类。CD-RBM无需人工标注情感词典,即可获得情感特征,且可以提高特征的情感语义关联性;同时,SVM可以保证产品评论情感分类的准确度。通过实验确定了RBM最优训练周期为10,在此训练周期下对RBM、SVM、PCA+SVM,以及RBM+SVM方法进行了比较。实验结果表明,RBM特征提取和SVM分类结合方法能够获得最好的准确率和F值,并获得较好的召回率。     

基于Hadoop架构的数据驱动的SVM并行增量学习算法

针对传统支持向量机（SVM）算法难以处理大规模训练数据的困境,提出一种基于Hadoop的数据驱动的并行增量Adaboost-SVM算法（PIASVM）。利用集成学习策略,局部分类器处理一个分区的数据,融合其分类结果得到组合分类器;增量学习中用权值刻画样本的空间分布特性,对样本进行迭代加权,利用遗忘因子实现新增样本的选择及历史样本的淘汰;采用基于HBase的控制器组件用以调度迭代过程,持久化中间结果并减小MapReduce原有框架迭代过程中的带宽压力。多组实验结果表明,所提算法具有优良的加速比、扩展率和数据伸缩度,在保证分类精度的基础上提高了SVM算法对大规模数据的处理能力。     

Modeling the correlations of crude oil properties based on sensitivity based linear learning method

This paper presented a new prediction model of pressure-volume-temperature (PVT) properties of crude oil systems using sensitivity based linear learning method (SBLLM). PVT properties are very important in the reservoir engineering computations. The accurate determination of these properties, such as bubble-point pressure and oil formation volume factor, is important in the primary and subsequent development of an oil field. Earlier developed models are confronted with several limitations especially their instability and inconsistency during predictions. In this paper, a sensitivity based linear learning method (SBLLM) prediction model for PVT properties is presented using three distinct databases while comparing forecasting performance, using several kinds of evaluation criteria and quality measures, with neural network and the three common empirical correlations. In the formulation used, sensitivity analysis coupled with a linear training algorithm for each of the two layers is employed which ensures that the learning curve stabilizes soon and behaves homogenously throughout the entire process operation. In this way, the model will be able to adequately model PVT properties faster with high stability and consistency. Empirical results from simulations demonstrated that the proposed SBLLM model produced good generalization performance, with high stability and consistency, which are requisites of good prediction models in reservoir characterization and modeling.     

Orthogonal sparse linear discriminant analysis

Linear discriminant analysis (LDA) is a linear feature extraction approach, and it has received much attention. On the basis of LDA, researchers have done a lot of research work on it, and many variant versions of LDA were proposed. However, the inherent problem of LDA cannot be solved very well by the variant methods. The major disadvantages of the classical LDA are as follows. First, it is sensitive to outliers and noises. Second, only the global discriminant structure is preserved, while the local discriminant information is ignored. In this paper, we present a new orthogonal sparse linear discriminant analysis (OSLDA) algorithm. The k nearest neighbour graph is first constructed to preserve the locality discriminant information of sample points. Then, L_2,1-norm constraint on the projection matrix is used to act as loss function, which can make the proposed method robust to outliers in data points. Extensive experiments have been performed on several standard public image databases, and the experiment results demonstrate the performance of the proposed OSLDA algorithm.     

A multi-class SVM classification system based on learning methods from indistinguishable chinese official documents

Support Vector Machines (SVM) has been developed for Chinese official document classification in One-against-All (OAA) multi-class scheme. Several data retrieving techniques including sentence segmentation, term weighting, and feature extraction are used in preprocess. We observe that most documents of which contents are indistinguishable make poor classification results. The traditional solution is to add misclassified documents to the training set in order to adjust classification rules. In this paper, indistinguishable documents are observed to be informative for strengthening prediction performance since their labels are predicted by the current model in low confidence. A general approach is proposed to utilize decision values in SVM to identify indistinguishable documents. Based on verified classification results and distinguishability of documents, four learning strategies that select certain documents to training sets are proposed to improve classification performance. Experiments report that indistinguishable documents are able to be identified in a high probability and are informative for learning strategies. Furthermore, LMID that adds both of misclassified documents and indistinguishable documents to training sets is the most effective learning strategy in SVM classification for large set of Chinese official documents in terms of computing efficiency and classification accuracy.     

A learning style classification mechanism for e-learning

With the growing demand in e-learning, numerous research works have been done to enhance teaching quality in e-learning environments. Among these studies, researchers have indicated that adaptive learning is a critical requirement for promoting the learning performance of students. Adaptive learning provides adaptive learning materials, learning strategies and/or courses according to a student’s learning style. Hence, the first step for achieving adaptive learning environments is to identify students’ learning styles. This paper proposes a learning style classification mechanism to classify and then identify students’ learning styles. The proposed mechanism improves k-nearest neighbor (k-NN) classification and combines it with genetic algorithms (GA). To demonstrate the viability of the proposed mechanism, the proposed mechanism is implemented on an open-learning management system. The learning behavioral features of 117 elementary school students are collected and then classified by the proposed mechanism. The experimental results indicate that the proposed classification mechanism can effectively classify and identify students’ learning styles.     

Early detection of cardiovascular autonomic neuropathy: A multi-class classification model based on feature selection and deep learning feature fusion

The conventional diagnostic process and tools of cardiovascular autonomic neuropathy (CAN) can easily identify the two main categories of the condition: severe/definite CAN and normal/healthy without CAN. Conventional techniques encounter significant challenges when identifying CAN in its early or atypical stages due to the inherent imbalanced and incompleteness condition in the collected clinical multimodal data, including electrocardiogram (ECG) data from ECG sensors, blood chemistry, podiatry, and endocrinology features. Therefore, most detection tools and techniques are limited to binary CAN classification. However, early diagnosis of CAN or diagnosis of the atypical stages of CAN is more important than the diagnosis of severe CAN, which, in fact, is easily identifiable with a few diagnostic reports. In this paper, we propose a novel multi-class classification approach for timely CAN detection. The proposed classification algorithm develops a multistage fusion model by combining feature selection and multimodal feature fusion techniques. The proposed method develops a performance criterion-based feature selection technique to guarantee highly significant features. A multimodal feature fusion technique was developed using deep learning feature fusion and selected original features. The experimental results obtained from testing with a large CAN dataset indicate that the proposed algorithm significantly improved the diagnostic accuracy of CAN compared to conventional Ewing battery features. The algorithm also identified the early or atypical stages of CAN with an AUC score of 0.931 using leave-one-out cross-validation.     

基于L1/2范数约束增量非负矩阵分解的SAR目标识别

增量非负矩阵分解(INMF)随目标样本增加逐渐更新分解模型,能够有效解决NMF算法的计算代价随样本增加而成倍增长的问题。然而INMF在使NMF具备增量学习能力的同时,并未考虑NMF分解矩阵的稀疏性对识别性能的提升作用。针对上述问题,提出基于L1/2范数约束的增量非负矩阵分解(L1/2-INMF)算法,并应用于SAR目标识别。L1/2-INMF采用L1/2范数实时约束增量过程中的NMF分解矩阵,能够在不增加计算复杂度的同时,提升识别性能。针对MSTAR数据集的仿真实验结果表明,提出的L1/2-INMF能够解决传统非负矩阵分解方法计算代价随样本增加而增加的问题。