一种概率映射网络的EM训练算法

文中提出一种概率映射网络(PMN)的EM(Expectation Maximization)训练算法。PMN为一个四层前馈网。它构成一个贝叶斯分类器,实现多类分类的贝叶斯判别,把输入的样本模式经网络变换为输出的分类判决,其网络节点对应于贝叶斯后验概率公式的各个变量。 此PMN用高斯核函数作为密度函数,网络参数训练由EM算法实现,其学习方式为类间的监督学习和类内的非监督学习。最后的实验表明此网络及其学习算法在分类应用中的有效性。     

基于隐变量贝叶斯模型的稀疏信号恢复

该文基于贝叶斯分析的视角,揭示了一类算法,包括使用隐变量模型的稀疏贝叶斯学习(SBL),正则化FOCUSS算法以及Log-Sum算法之间的内在关联。分析显示,作为隐变量贝叶斯模型的一种,稀疏贝叶斯学习使用第2类最大似然(Type II ML)在隐变量空间进行运算,可以视作一种更为广义和灵活的方法,并且为不适定反问题的稀疏求解提供了改进的途径。较之于目前基于第1类最大似然(Type I ML)的稀疏方法,仿真实验证实了稀疏贝叶斯学习的优越性能。     

融合关联规则与知识的贝叶斯网络学习算法

领域知识可以有效的提高贝叶斯网络学习效率与精度.文中提出了基于关联规则的SEM算法——AR-SEM算法.AR-SEM算法首先利用关联规则分析变量间的因果关系,并作为初始先验知识和领域专家的意见相结合,进一步去除无意义的规则,形成一个知识库,最后将知识库与SEM算法相结合来构造贝叶斯网络.文中在具有一定缺省数据的数据集上进行实验,实验表明AR-SEM可有效提高贝叶斯网络结构学习的精度.     

基于遗传算法的贝叶斯网络结构学习综述

贝叶斯网络采用图模型描述变量之间的依赖关系，因其结构清晰，具有突出的决策机制和学习机制，故拥有优秀的推理能力。在各类研究方法中，遗传算法能够有效地解决复杂的优化问题，以其普适性好、鲁棒性强、便于并行执行、高效便捷等显著特点，在贝叶斯网络结构的学习研究过程中发挥着非常重要的作用。从初始种群、遗传操作算子设计两个层面对近年基于遗传算法的因果结构学习改进方法进行了调研分析并指出了该技术路线进一步的研究方向。     

基于遗传算法的NoC路径分配算法

在片上网络中实现通信流明确的应用,通常在编译过程中静态分配路径资源,并把路径分配算法嵌入到映射算法中综合考虑.针对现有基于遗传算法的片上网络路径分配算法,引入了一种完整路径均匀交叉算子,来改善现有算法中路径交叉不充分的问题.实验结果显示:使用新算子的路径分配算法优化了现有算法的结果,减少了计算时间.     

基于混合方式的贝叶斯网络结构学习

基于最大主子图分解技术和遗传算法,提出了一种混合方式的贝叶斯网络结构学习算法。该算法首先根据领域知识和观察数据构造网络的无向独立图,并对其进行最大主子图分解,再利用遗传算法学习每个子图的结构,同时进行合并修正得到最优的贝叶斯网络结构。分解过程将一个学习大网络问题转化为小子图的学习问题,降低了搜索空间。仿真结果表明,新算法的学习效果与运行效率均有明显提高。     

基于数据挖掘技术的网络热门舆情分类研究

以提升网络热门舆情分类准确率,降低分类时间为目标,提出了基于数据挖掘技术的网络热门舆情分类方法.将小波核函数和支持向量机结合构成小波模糊支持向量机,采用增量学习机制和贝叶斯分类算法建立增量贝叶斯分类算法,组成小波模糊支持向量机-增量贝叶斯分类算法解决测试样本易分类失误以及类条件独立假定性很难获取问题,通过计算待测样本和...     

生物缺失数据处理的贝叶斯模型研究

文章针对生物信息实验中的分类预测问题,以属性缺失数据为对象,结合朴素贝叶斯算法的特点,设计了一种基于改进EM算法的缺失数据朴素贝叶斯填充模型,并应用于蛋白质作用位点的定位研究中.实验结果表明,通过算法进行生物缺失数据的处理,在准确率、精度、召回率、ROC方面均获得了比其他方法更好的效果.     

用于入侵的自适应遗传算法训练人工神经网络

给出了一种能和网络结构一一对应的、合适的染色体编码方法.用物种入侵的遗传算法训练人工神经网络,在入侵过程中,遗传算法自适应地调整交叉算子和变异算子.提出了一种根据平均适应度值确定入侵物种规模的方法,并详细描述了算法步骤,最后通过实验证明了本文算法的有效性和优越性.     

基于BD度量的简单贝叶斯优化学习算法

贝叶斯优化算法是利用贝叶斯网络匹配进化种群的优良解集而产生新的染色体来体现种群的进化.在贝叶斯网络对种群进行匹配的过程中,贝叶斯网络结构越复杂,种群的进化信息描述越完整,进化质量越高,但运算速度相对来说越慢;相反,贝叶斯网络越简单,算法描述的种群的进化信息越少,进化质量越差,但却能够提高算法的运算速度.基于此,给出了简单贝叶斯优化与复杂贝叶斯优化定义.针对简单贝叶斯网络提出了基于BD度量的三步结构学习算法,并给出了一个利用这种算法进行贝叶斯网络结构学习的例子.     

基于遗传算法的动态Bayesian网结构学习的研究

动态Bayesian网是复杂随机过程的图形表示形式,从数据中学习建造动态Bayesian网是目前的研究热点问题.本文针对该问题提出了一种遗传算法.文中设计了结合数学期望的适应度函数,该函数利用进化过程中的最好动态Bayesian网把不完备数据转换成完备数据,使动态Bayesian网的学习分解为两个Bayesian网(初始网和转换网)的学习,简化了学习的复杂度.此外,文中给出了网络结构的编码方案,设计了相应的遗传算子.模拟实验结果表明,该算法能有效地从不完备数据序列中学习动态Bayesian网,并且实验结果说明了隐藏变量的作用和遗传控制参数对结果模型的影响.     

Optimised competitive feature vector network

A new, competitive, incremental network architecture is compared with learning vector quantisation (LVQ) and grow and learn (GAL) networks. Optimisation of the feature vectors in the new algorithm is currently implemented by a genetic algorithm (GA)     

多贝叶斯网络分类器集成模型研究

提出了一种多贝叶斯网络集成的分类和预测方法.把专家知识作为"疫苗",利用免疫遗传算法和约束信息熵适应度函数相结合的方法进行贝叶斯网络结构的学习,得到多个反映同一样本数据集的、网络结构复杂度折衷的、满意的贝叶斯网络结构.然后,给出了多贝叶斯网络分类器集成模型,把学习得到的贝叶斯网络进行集成,代表"专家"对未知类别的不完全数据进行群决策的分类和预测,提升贝叶斯网络分类器的泛化能力.最后,结合贝叶斯推理工具GeNIe软件,通过实例说明该方法的合理性和有效性.     

Structural Learning of Bayesian Networks using a modified MDL score metric

Bayesian networks are tools as they represent probability distributions as graphs. They work with uncertainties of real systems. Since last decade there is a special interest in learning network structures from data. However learning the best network structure is a NP-Hard problem, so many heuristics algorithms to generate network structures from data were created. Many of these algorithms use score metrics to generate the network model. This paper address learn the structure of ALARM pattern benchmark using K-2 algorithm and a modified MDL as score metric. Results shown that score metrics with parameters that strength the tendency to select simpler network structures are better than score metrics with weaker tendency to select simpler network structures and that modified MDL gives better results than original MDL.     

Defense decision-making method based on incomplete information stochastic game and Q-learning

Most of the existing stochastic games are based on the assumption of complete information,which are not consistent with the fact of network attack and defense.Aiming at this problem,the uncertainty of the attacker’s revenue was transformed to the uncertainty of the attacker type,and then a stochastic game model with incomplete information was constructed.The probability of network state transition is difficult to determine,which makes it impossible to determine the parameter needed to solve the equilibrium.Aiming at this problem,the Q-learning was introduced into stochastic game,which allowed defender to get the relevant parameter by learning in network attack and defense and to solve Bayesian Nash equilibrium.Based on the above,a defense decision algorithm that could learn online was designed.The simulation experiment proves the effectiveness of the proposed method.     

An incremental learning method based on probabilistic neural networks and adjustable fuzzy clustering for human activity recognition by using wearable sensors

Human activity recognition by using wearable sensors has gained tremendous interest in recent years among a range of health-related areas. To automatically recognize various human activities from wearable sensor data, many classification methods have been tried in prior studies, but most of them lack the incremental learning abilities. In this study, an incremental learning method is proposed for sensor-based human activity recognition. The proposed method is designed based on probabilistic neural networks and an adjustable fuzzy clustering algorithm. The proposed method may achieve the following features. 1) It can easily learn additional information from new training data to improve the recognition accuracy. 2) It can freely add new activities to be detected, as well as remove existing activities. 3) The updating process from new training data does not require previously used training data. An experiment was performed to collect realistic wearable sensor data from a range of activities of daily life. The experimental results showed that the proposed method achieved a good tradeoff between incremental learning ability and the recognition accuracy. The experimental results from comparison with other classification methods demonstrated the effectiveness of the proposed method further.     

一种基于改进教学优化的微弱信号检测方法

为了快速准确地检测混沌背景中的微弱信号,提高网络泛化能力,文中利用改进教学优化算法优化贝叶斯回声状态网络的模型参数,提出了一种改进教学优化的混沌背景中微弱信号检测方法。通过建立混沌序列单步预测模型,分析预测误差的幅值,检测混沌背景中微弱瞬态信号和周期信号。对Lorenz系统和实测的海杂波数据进行实验研究,验证预测模型的有效性,结果表明,贝叶斯回声状态网络模型的预测结果比支持向量机和径向基神经网络模型的均方根误差降低了2个数量级,缩短了预测时间,提高了预测精度和预测效率,能快速有效地检测混沌背景中微弱信号,且具有更低的门限。     

Network Inference From Co-Occurrences

The discovery of networks is a fundamental problem arising in numerous fields of science and technology, including communication systems, biology, sociology, and neuroscience. Unfortunately, it is often difficult, or impossible, to obtain data that directly reveal network structure, and so one must infer a network from incomplete data. This paper considers inferring network structure from “co-occurrence” data: observations that identify which network components (e.g., switches, routers, genes) carry each transmission but do not indicate the order in which they handle the transmission. Without order information, the number of networks that are consistent with the data grows exponentially with the size of the network (i.e., the number of nodes). Yet, the basic engineering/evolutionary principles underlying most networks strongly suggest that not all data-consistent networks are equally likely. In particular, nodes that co-occur in many observations are probably closely connected. With this in mind, we model the co-occurrence observations as independent realizations of a random walk on the network, subjected to a random permutation to account for the lack of order information. Treating permutations as missing data, we derive an expectation–maximization (EM) algorithm for estimating the random walk parameters. The model and EM algorithm significantly simplify the problem, but the computational complexity of the reconstruction process does grow exponentially in the length of each transmission path. For networks with long paths, the exact E-step may be computationally intractable. We propose a polynomial-time Monte Carlo EM algorithm based on importance sampling and derive conditions that ensure convergence of the algorithm with high probability. Simulations and experiments with Internet measurements demonstrate the promise of this approach.