结构正则化多视图非负矩阵分解

现存非负矩阵分解（non-negative matrix factorization,NMF）研究多考虑单一视图分解数据,忽略了数据信息的全面性。此外,NMF限制其获取数据的内在几何结构。针对以上问题,提出一个结构正则化多视图非负矩阵分解算法（structure regularized multi-view nonnegative matrix factorization,SRMNMF）。首先,通过主成分分析来对数据进行全局结构的判别式学习;其次,利用流形学习来捕获数据的局部结构;然后,通过利用多视图数据的多样性和差异性来学习表征。模型提升了算法聚类的整体性能,更加有效地挖掘数据的结构信息。此外,采用高效的交替迭代算法优化目标函数得到最优的因子矩阵。在六个数据集上与现存的代表性方法比较,所提出的SRMNMF的准确率、NMI和Purity分别最大提高4.4%、6.1%和4.05%。     

三峡水库运行后宜昌-城陵矶河段冲刷重心下移与时空演变

三峡大坝下游河道的冲刷发展影响着长江中下游的防洪、航运及区域经济发展。本文以宜昌至城陵矶河段约400 km长河段为研究对象,基于2002—2018年河道的来水来沙、泥沙粒径和横断面形态观测资料,计算并分析河道垂向冲淤、横向摆动、平滩面积、平滩水深和冲淤量的变化,根据河道平面形态将研究河段划分为32个子河段,对子河段平均平滩面积和平滩水深的变化速率进行分级,引入空间聚类分析方法,对冲刷强度等级最高的河段在时空尺度上分布进行聚类分析,研究河道冲刷重心的时空迁移规律。结果表明,三峡水库运用后下游河道经历了蓄水初期冲刷较强（2003—2007）、175 m正常蓄水运用后冲刷减弱（2008—2012）和上游梯级水库运用后冲刷增强（2013—2018）的三个阶段;空间上,宜昌至枝城及上荆江河段普遍冲刷,以弯曲型为主的下荆江河道以冲为主,但冲淤交替频繁,同时冲淤强度具有随时间衰减的特点;冲刷重心具有向下游迁移的趋势,在三峡水库蓄水初期冲刷重心向下迁移较快,之后减缓,上游梯级水库运用后,冲刷重心下移速率明显加快,之后再次减缓,2003—2018年冲刷重心平均下移速率约7.5 km/a。     

Potential applications of three-way multidimensional scaling and related techniques to integrate knowledge from multiple experts

Theknowledge transfer problem in artificial intelligence consists of finding effective ways to elicit information from a human expert and represent it in a form suitable for use by an expert system. One approach to formalizing and guiding this knowledge transfer process for certain types of expert systems is to use psychometric scaling methods to analyze data on how the human expert compares or groups solutions. For example, Butler and Corter [1] obtained judgments of thesubstitutability of solutions from an expert, then analyzed the resulting data via techniques for fitting trees and extended trees [2]. The expert's interpretation of certain aspects of the solutions were directly encoded as production rules, allowing rapid prototyping. In this paper we consider the problem of combining information from multiple experts. We propose the use of three-way or individual differences multidimensional scaling, tree-fitting, and unfolding models to analyze two types of data obtainable from the multiple experts: judgments of the substitutability of pairs of solutions, and judgments of the appropriateness of specific solutions to specific problems. An application is described in which substitutability data were obtained from three experts and analyzed using the SINDSCAL program [3] for three-way multidimensional scaling [4].     

深度学习在视频对象分割中的应用与展望

视频对象分割是指在给定的一段视频序列的各帧图像中,找出属于特定前景对象的所有像素点位置区域.随着硬件平台计算能力的提升,深度学习受到了越来越多的关注,在视频对象分割领域也取得了一定的进展.本文首先介绍了视频对象分割的主要任务,并总结了该任务所面临的挑战.其次,对开放的视频对象分割常用数据集进行了简要概述,并介绍了通用的性能评估标准.接着,综述了视频对象分割的研究现状,详细地分析了当前的各种方法,并将它们划分为三大类:半监督的方法,即给出视频第一帧图像中感兴趣对象的详细人工真值标注,分割出视频剩余图像中的感兴趣对象;无监督的方法,即不给任何人工标注信息,自动识别并分割出视频中的前景对象;交互式的方法,即在分割过程中,通过人工交互式的参与,结合粗略的人工标注先验信息,进行视频对象分割.第三类方法的条件相当于前两者的折中:相对于第一类方法,它虽然需要人工的参与,但只需要少量的标注工作量;相对于第二类方法,它给视频序列中某些帧的图像适当地添加了一些人工标注信息,从而更具针对性.最后,对深度学习在视频对象分割任务中的应用,进行了总结和展望.     

An intelligent hybrid approach for hepatitis disease diagnosis: Combining enhanced k‐means clustering and improved ensemble learning

In real world, the automatic detection of liver disease is a challenging problem among medical practitioners. The intent of this work is to propose an intelligent hybrid approach for the diagnosis of hepatitis disease. The diagnosis is performed with the combination of k‐means clustering and improved ensemble‐driven learning. To avoid clinical experience and to reduce the evaluation time, ensemble learning is deployed, which constructs a set of hypotheses by using multiple learners to solve a liver disease problem. The performance analysis of the proposed integrated hybrid system is compared in terms of accuracy, true positive rate, precision, f‐measure, kappa statistic, mean absolute error, and root mean squared error. Simulation results showed that the enhanced k‐means clustering and improved ensemble learning with enhanced adaptive boosting, bagged decision tree, and J48 decision tree‐based intelligent hybrid approach achieved better prediction outcomes than other existing individual and integrated methods.     

一种半监督学习的代码自动生成性能评估方法

为了提高软件开发的质量和效率,代码自动生成是当前的研究热点,代码自动生成的性能是其中的重要问题.现有代码自动生成的性能分析方法较简单,难以评估代码自动生成过程中程序员与代码自动生成工具各自的特征.本文综合考虑了代码自动生成过程中程序员与代码自动生成工具的作用,提出了一种基于半监督学习的代码自动生成性能评估方法,通过抽取程序员行为与代码自动生成工具行为的重要特征,划分代码自动生成的性能类别,建立了基于深度神经网络的代码自动生成过程性能评估模型,并计算程序员行为特征与代码自动生成工具行为特征对性能的影响程度.实验结果表明,该方法可以有效分析程序员行为与代码自动生成工具行为对代码自动生成过程性能的影响.     

基于半监督学习和支持向量机的煤与瓦斯突出预测研究

针对支持向量机要求输入向量为已标记样本,而实际应用中已标记样本很难获取的问题,提出将半监督学习和支持向量机结合的煤与瓦斯突出预测方法;介绍了采用SVM预测煤与瓦斯突出的流程及其输入向量的选择;对半监督学习中的协同训练算法进行了改进:在同一属性集上训练2个不同分类器SVM和KNN,将2个分类器标记一致的样本加入训练集,从而充分利用未标记样本不断补充信息,更新训练集标记样本,达到强化训练集的目的。测试结果表明,改进后的算法比单独的支持向量机预测方法准确率更高。     

基于能量强度的簇区分路由方法研究

传感器节点体积微小,通常携带能量十分有限的电池。由于传感器节点个数多,成本要求低廉,分布区域广,而且部署区域环境复杂,有些区域甚至是人员不能到达的,所以传感器节点通过更换电池的方式来补充能量是不现实的。如何高效使用能量来最大化网路的生命周期是无线传感器网络面临的首要挑战。为了提高网络的寿命,目前人们已经提出了很多基于网络层的路由协议。比较研究了传感器网络现有的路由协议,参考LEACH分簇算法和MTP协议,提出了一种在MTP协议基础上将LEACH分簇思想引入进来的改进方法,即在基于能量强度将网络节点分层后,再在各层上进行簇区分,均衡各传感器节点的能量来提高网络的寿命。实验表明该路由算法可提高网络寿命。     

A new ECG beat clustering method based on kernelized fuzzy c-means and hybrid ant colony optimization for continuous domains

The kernelized fuzzy c-means algorithm uses kernel methods to improve the clustering performance of the well known fuzzy c-means algorithm by mapping a given dataset into a higher dimensional space non-linearly. Thus, the newly obtained dataset is more likely to be linearly seprable. However, to further improve the clustering performance, an optimization method is required to overcome the drawbacks of the traditional algorithms such as, sensitivity to initialization, trapping into local minima and lack of prior knowledge for optimum paramaters of the kernel functions. In this paper, to overcome these drawbacks, a new clustering method based on kernelized fuzzy c-means algorithm and a recently proposed ant based optimization algorithm, hybrid ant colony optimization for continuous domains, is proposed. The proposed method is applied to a dataset which is obtained from MIT–BIH arrhythmia database. The dataset consists of six types of ECG beats including, Normal Beat (N), Premature Ventricular Contraction (PVC), Fusion of Ventricular and Normal Beat (F), Artrial Premature Beat (A), Right Bundle Branch Block Beat (R) and Fusion of Paced and Normal Beat (f). Four time domain features are extracted for each beat type and training and test sets are formed. After several experiments it is observed that the proposed method outperforms the traditional fuzzy c-means and kernelized fuzzy c-means algorithms.