A novel temporal error concealment method based on fuzzy reasoning for H.264

In this paper, a fuzzy reasoning based temporal error concealment method is proposed. The basic temporal error concealment is implemented by estimating Motion Vector (MV) of the lost MacroBlock (MB) from its neighboring MVs. Which MV is the most proper one is evaluated by some criteria. Generally, two criteria are widely used, namely Side Match Distortion (SMD) and Sum of Absolute Difference (SAD) of corresponding MV. However, each criterion could only partly describe the status of lost block. To accomplish the judgement more accurately, the two measures are considered together. Thus a refined measure based on fuzzy reasoning is adopted to balance the effects of SMD and SAD. Terms SMD and SAD are regarded as fuzzy input and the term ‘similarity’ as output to complete fuzzy reasoning. Result of fuzzy reasoning repre-sents how the tested MV is similar to the original one. And k-means clustering technique is performed to de-fine the membership function of input fuzzy sets adaptively. According to the experimental results, the con-cealment based on new measure achieves better performance.     

数据挖掘算法性能优化的研究与应用

数据挖掘中经常使用k-means算法,它是经常使用的一种聚类分析算法,但易受初始聚类中心和聚类个数k的影响。因此对近年从算法原理、关键技术和优缺点等方面提出的较有代表性的关于初始聚类中心和k值确定的改进的k-means算法进行了分析。并选用知名数据集对一些典型算法进行测试和应用。上述工作将为数据挖掘的研究提供有益的参考。     

基于新径向基函数网络的变压器故障诊断法

油中溶解气体分析(DGA)是判别变压器内部绝缘状况及发现内部潜伏性故障的重要手段,而多层前馈网络(MLFNN)是应用广泛的故障诊断模型。为此,提出了以DGA数据为特征参数的新型径向基函数神经网络(RBFNN)诊断变压器故障。在分析传统k-均值聚类算法RBFNN的缺点和最优聚类特性的基础上,介绍了RBFNN的新算法-自适应k-均值聚类算法,它既能避免传统k-均值聚类算法的局部收敛的缺点,又能动态调整学习率。最后,大量聚类实验结果显示自适应k-均值聚类算法在收敛速度和聚类性能上比传统k-均值聚类算法显著提高;故障诊断实验结果显示所提出的模型故障诊断准确度高于传统BPNN、RBFNN及IEC三比值法。     

Approximate computing method based on cross-layer dynamic precision scaling for the k-means

With the application of artificial intelligence on the embedded platform, the k-means clustering algorithm, as the basis of the artificial intelligence method, is implemented on the embedded platform. Energy consumption is the key for the algorithm implementation on the embedded platform. In order to reduce the energy consumption of the k-means on the embedded platform, an approximate computing method based on cross-layer dynamic precision scaling for the k-means is proposed. First, the iteration process is constrained from the distance between data point to centroid and data point change trend. And a dynamic precision scaling method is proposed. Then the data reorganization and access method of external memory is designed from the structural level, which can realize the access of approximate memory. In addition, the approximate adder and multiplier are designed which can automatically adjust the calculation accuracy. Finally, the approximate computing of the k-means is realized. Experimental results show that the proposed method can reduce the energy consumption by 55%~58% compared with the accurate computing without affecting the quality of clustering. The proportion of the energy saving is the highest.     

Early COVID-19 Symptoms Identification Using Hybrid Unsupervised Machine Learning Techniques

The COVID-19 virus exhibits pneumonia-like symptoms, including fever, cough, and shortness of breath, and may be fatal. Many COVID-19 contraction experiments require comprehensive clinical procedures at medical facilities. Clinical studies help to make a correct diagnosis of COVID-19, where the disease has already spread to the organs in most cases. Prompt and early diagnosis is indispensable for providing patients with the possibility of early clinical diagnosis and slowing down the disease spread. Therefore, clinical investigations in patients with COVID-19 have revealed distinct patterns of breathing relative to other diseases such as flu and cold, which are worth investigating. Current supervised Machine Learning (ML) based techniques mostly investigate clinical reports such as X-Rays and Computerized Tomography (CT) for disease detection. This strategy relies on a larger clinical dataset and does not focus on early symptom identification. Towards this end, an innovative hybrid unsupervised ML technique is introduced to uncover the probability of COVID-19 occurrence based on the breathing patterns and commonly reported symptoms, fever, and cough. Specifically, various metrics, including body temperature, breathing and cough patterns, and physical activity, were considered in this study. Finally, a lightweight ML algorithm based on the K-Means and Isolation Forest technique was implemented on relatively small data including 40 individuals. The proposed technique shows an outlier detection with an accuracy of 89%, on average.     

Improved Pattern Clustering Algorithm and its Application in Recognition of Transversal Thickness Distribution of Steel Strip

 Transversal thickness distribution of steel strip in the entry section of cold rolling mill has distinct affections to the flatness and transversal thickness control precision of final products. Pattern clustering method is introduced to steel rolling area and is first time to be used in the patterns recognition of transversal thickness distribution of steel strip. K-means clustering algorithm as the best-known one has the advantage of being easy to implement, it has drawbacks. In this paper, an improved k-means clustering algorithm is presented, main improvement points include the amount of clusters is indirectly determined by experience, the initial clustering points are preselected according to the density queue of data objects and Mahalanobis distance is applied instead of Euclidean distance. Compared to the clustered patterns obtained from the common k-means algorithm, the patterns identified from the improved algorithm is more reasonable. The results of application in one coil further show the improved clustering algorithm is well suitable for the patterns’ recognition of transversal thickness distribution of steel strip. It will do great help in the online quality control system.     

社交网络重叠社团检测初探磁

大数据环境下社交网络的社团结构研究对解决很多现实问题有着重要的意义。社团通常被看作是有相对紧密的内部连接和比较稀疏的外部连接的子图,重叠是社团结构的一个重要特征。论文基于 G（n ,p）模型,提出了一种生成包含重叠社团的合成网络的方法,然后基于 k-均值和随机游走设计了一种重叠社团的检测算法,并在合成网络上初步验证了该方法的可行性。     

基于k-d树的k-means聚类方法

在直接k-means算法的基础上提出了一种新的基于k-d树的聚类方法。通过把所有的对象组织在一棵k-d树中,可以高效地发现给定原型的所有最近邻对象。利用的主要思想是：在根结点,所有的聚类中心(或称为候选原型)都是所有对象的最近邻候选集合,对于根结点的子结点,通过简单几何约束来剪枝该候选集,这种方法可以被递归使用。使用基于k-d树的方法可以使直接k-means算法的总体性能提高一到两个数量级。     

改进的基于遗传算法的粗糙聚类方法

传统的聚类算法都是使用硬计算来对数据对象进行划分,然而现实中不同类之间对象通常没有明确的界限。粗糙集理论提供了一种处理边界对象不确定的方法。因此将粗糙理论与k-均值方法相结合。同时,传统的k-均值聚类方法必须事先给定聚类数k,但实际情况下k很难确定;另外虽然传统k-均值算法局部搜索能力强,但容易陷入局部最优。遗传算法能得到全局最优解,但收敛过快。鉴于此,提出了一种改进的基于遗传算法的的粗糙聚类方法。该算法能动态地生成k-均值聚类数,采用最大最小原则生成初始聚类中心,同时结合粗糙集理论的上近似和下近似处理边界对象。最后,用UCI的Iris数据集分别对算法进行实际验证。实验结果表明,该算法具有较高的正确率,综合性能更加稳定。     

K-means聚类和支持向量机结合的文本分类研究

针对数据挖掘中文本自动分类问题,提出了一种基于k-means聚类算法和支持向量机相结合的文本分类方法。该方法先将文本大致聚为k类,然后对每一类用支持向量机进行细分。构造了可用于多个模式类识别的多层SVM模型,该模型可完成对多个模式的分类识别。给出了该模型的构造及应用的方法,并验证了该方法的有效性。