包含隐变量的贝叶斯网络增量学习方法

提出了一种贝叶斯网络增量学习方法——ILBN.ILBN将EM算法和遗传算法引入到了贝叶斯网络的增量学习过程中,用EM算法从不完整数据计算充分统计量的期望,用遗传算法进化贝叶斯网络的结构,在一定程度上缓解了确定性搜索算法的局部极值问题.通过定义新变异算子和扩展传统的交叉算子,ILBN能够增量学习包含隐变量的贝叶斯网络结构.最后,ILBN改进了Friedman等人的增量学习过程.实验结果表明,ILBN和Friedman等人的增量学习方法存储开销相当,但在相同条件下,学到的网络更精确;实验结果也证实了存在不完整数据和隐变量时,ILBN的增量学习能力.     

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.     

基于改进鲸鱼优化策略的贝叶斯网络结构学习算法

针对当前贝叶斯网络结构学习算法易陷入局部最优和寻优效率低的问题,该文提出一种基于改进鲸鱼优化策略的贝叶斯网络结构学习算法。该算法首先提出一种新的方法建立较优的初始种群,然后利用不产生非法结构的交叉变异算子构建适用于贝叶斯网络结构学习的改进捕食行为,同时采用动态调节参数增强算法个体寻优的能力,通过适应度排序更新种群,最终获得最优的贝叶斯网络结构。仿真结果表明,该算法具有全局收敛性,寻优效率高,精确率高于其它同类优化算法。     

一种基于模型的可分解贝叶斯在线强化学习

针对贝叶斯强化学习中参数个数巨大,收敛速度慢,无法实现在线学习的问题,提出一种基于模型的可分解贝叶斯强化学习方法.首先,将学习参数进行可分解表示,降低学习参数的个数;然后,根据先验知识和观察数据采用贝叶斯方法来学习,最优化探索和利用二者之间的平衡关系;最后,采用基于点的贝叶斯强化学习方法实现学习过程的快速收敛,从而达到在线学习的目的.仿真结果表明该算法能够满足实时系统性能的要求.     

基于MDL原理与混合遗传算法的Bayesian网络结构学习

从大型数据库中学习Bayesian网络结构是Bayesian网络应用的难点之一。在分析标准遗传算法与爬山算法各自优点与不足的基础上，将这两种算法相结合，以最小描述长度为评价函 数，得到一种混合遗传算法，实现了它们的优势互补。文章给出了混合遗传算法的计算步骤，并通过对ALARM数据库学习得到的Bayesian网络结构。     

基于马尔科夫毯约束的最优贝叶斯网络结构学习算法

本文针对最优贝叶斯网络的结构学习问题,在动态规划算法（Dynamic Programming,DP）的基础上,使用IAMB算法（Incremental Association Markov Blanket,IAMB）计算得到的马尔科夫毯对评分计算过程进行约束,减少了评分的计算次数,提出了基于马尔科夫毯约束的动态规划算法（Dynamic Programming Constrained with Markov Blanket,DPCMB）,研究了IAMB算法中重要性阈值对DPCMB算法的各项性能指标的影响,给出了调整阈值的合理建议.实验结果表明,DPCMB算法可以通过调整重要性阈值,使该算法的精度与DP算法相当,极大地减少了算法的运行时间、评分计算次数和所需存储空间.     

DOA Estimation Using Block Variational Sparse Bayesian Learning

In Direction-of-arrival (DOA) estimation, the real-valued sparse Bayesian algorithm degrades the es-timation performance by decomposing the complex value into real and imaginary components and combining them independently. We directly use complex probability density functions to model the noise and complex-valued sparse direction weights. Based on the Multiple measurement vectors (MMV), block sparse structure for the direction weights is integrated into the variational Bayesian learning to provide accurate source direction estimates. The pro-posed algorithm can be used for arbitrary array geome-tries and does not need the prior information of the in-cident signal number. Simulation results demonstrate the better performance of the proposed method compared with the real-valued sparse Bayesian algorithm, the Orthogo-nal matching pursuit (OMP) and l1 norm based complex-valued methods.     

Hyperparameter Optimization for Machine Learning Models Based on Bayesian Optimization

Hyperparameters are important for machine learning algorithms since they directly control the behaviors of training algorithms and have a significant effect on the performance of machine learning models. Several techniques have been developed and successfully applied for certain application domains. However, this work demands professional knowledge and expert experience. And sometimes it has to resort to the brute-force search. Therefore, if an efficient hyperparameter optimization algorithm can be developed to optimize any given machine learning method, it will greatly improve the efficiency of machine learning. In this paper, we consider building the relationship between the performance of the machine learning models and their hyperparameters by Gaussian processes. In this way, the hyperparameter tuning problem can be abstracted as an optimization problem and Bayesian optimization is used to solve the problem. Bayesian optimization is based on the Bayesian theorem. It sets a prior over the optimization function and gathers the information from the previous sample to update the posterior of the optimization function. A utility function selects the next sample point to maximize the optimization function. Several experiments were conducted on standard test datasets. Experiment results show that the proposed method can find the best hyperparameters for the widely used machine learning models, such as the random forest algorithm and the neural networks, even multi-grained cascade forest under the consideration of time cost.     

Enhancing Channel Estimation in Cognitive Radio Systems by means of Bayesian Networks

This paper proposes enhancements to the channel(-state) estimation phase of a cognitive radio system. Cognitive radio devices have the ability to dynamically select their operating configurations, based on environment aspects, goals, profiles, preferences etc. The proposed method aims at evaluating the various candidate configurations that a cognitive transmitter may operate in, by associating a capability e.g., achievable bit-rate, with each of these configurations. It takes into account calculations of channel capacity provided by channel-state estimation information (CSI) and the sensed environment, and at the same time increases the certainty about the configuration evaluations by considering past experience and knowledge through the use of Bayesian networks. Results from comprehensive scenarios show the impact of our method on the behaviour of cognitive radio systems, whereas potential application and future work are identified.

基于贝叶斯网络工具箱的贝叶斯学习和推理

采用MATLAB语言编制的贝叶斯网络工具箱(Bayesian Networks Toolbox,BNT)可实现贝叶斯网络结构学习、参数学习、推理和构建贝叶斯分类器,此工具箱在贝叶斯学习编程方面非常灵活.介绍了用贝叶斯网络工具箱解决贝叶斯学习和推理问题,并给出了两个实例.     

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.     

Learning Bayesian Networks with the Combination of Evolutionary Algorithm and EMI Method

Most Bayesian network （BN） learning algorithms use EMI algorithm to deal with incomplete data. But EMI algorithm is of low efficiency due to its iterative parameter refinement, and the problem will become even worse when multiple runs of EMI algorithm are needed. Besides, EMI algorithm usually converges to local maxima, which also degrades the accuracy of EMI based BN learning algorithms. In this paper, we replace EMI algorithm used in BN learning tasks with EMI method to deal with incomplete data. EMI is a very efficient method, which estimates probability distributions directly from incomplete data rather than performs iterative refinement of parameters. Base on EMI method, we propose an effec- tive algorithm, namely EMI-EA. EMI-EA algorithm uses EMI method to estimate probability distribution over local structures in BNs, and evaluates BN structures with a variant of MDL scoring function. To avoid getting into local maxima of the search process, EMI-EA evolves BN struc- tures with an Evolutionary algorithm （EA）. The experi- mental results on Alarm, Asia and an examplar network show that EMI-EA algorithm outperforms EMI-EA for all samples and E-TPDA algorithms for small and middle size of samples in terms of accuracy. In terms of efficiency, EMI-EA is comparable with E-TPDA algorithm and much more efficient than EMI-EA algorithm. EMI-EA also out- performs EMI-EA and M-V algorithm when learning BNs with hidden variables.     

一种通过结构边界进行贝叶斯网络学习的算法

贝叶斯网络是智能算法领域重要的理论工具,其结构学习问题被认为是NP-hard问题。该文通过混合学习算法的方式,从分析低阶条件独立性测试提供的信息入手,给出了构造目标网络结构空间边界的方法,并给出了完整的证明。在此基础上执行打分搜索算法获得最终的网络结构。仿真结果表明该算法与同类算法相比具有更高的精度和更好的执行效率。     

贝叶斯网络应用中的结构学习方法研究

贝叶斯网络是用于表示不确定变量之间潜在依赖关系的图形模型。结构学习是贝叶斯网络学习的核心,有效的结构学习方法和算法是构建最优网络结构的基础。文中对迄今为止贝叶斯网络应用中的结构学习方法进行探讨,从复杂度、适用性等方面对其进行分析比较,并指出每种方法的关键环节和主要思想,对实际应用中的方法选择和研究提供了参考。     

应用精英反向学习的多目标烟花爆炸算法

现实中的多目标优化问题越来越多,而且日益复杂.受混合多目标优化算法设计思想的启发,将烟花爆炸方法和精英反向学习机制引入至多目标优化领域,提出一种应用精英反向学习的多目标烟花爆炸算法(Multi-Objective Fireworks Optimization Algorithm Using Elite Opposition-Based Learning,MOFAEOL).该算法利用精英反向学习策略加强算法的全局搜索能力,利用烟花爆炸方法增强算法的局部搜索能力并提高求解的精度.这两种搜索机制相互协同以更好地平衡算法的全局勘探和局部开采的能力.MOFAEOL算法与另外5种代表性多目标优化算法一同在由ZDT系列和DTLZ系列组成的测试集上进行性能比较.实验表明,MOFAEOL算法在收敛性、多样性和稳定性方面均优于或部分优于其他对比算法.     

Wavelength Converters Placement in All Optical Networks Using Particle Swarm Optimization

Placement of wavelength converters in an arbitrary mesh network is known to be a NP-complete problem. So far, this problem has been solved by heuristic strategies or by the application of optimization tools such as genetic algorithms. In this paper, we introduce a novel evolutionary algorithm: particle swarm optimization (PSO) to find the optimal solution to the converters placement problem. The major advantage of this algorithm is that does not need to build up a search tree or to create auxiliary graphs in find the optimal solutions. In addition, the computed results show that only a few particles are needed to search the optimal solutions of the placement of wavelength converters problem in an arbitrary network. Experiments have been conducted to demonstrate the effectiveness and efficiency of the proposed evolutionary algorithm. It was found that the efficiency of PSO can even exceed 90% under certain circumstances. In order to further improve the efficiency in obtaining the optimal solutions, four strategic initialization schemes are investigated and compared with the random initializations of PSO particles.     

基于贝叶斯网络的通信网告警相关性和故障诊断模型

该文采用贝叶斯网络建立告警相关性和故障诊断模型。首先介绍了基于贝叶斯网络推理的基本概念。提出了通信网功能分层结构的思想,建立不同网络层次间的故障传播模型。详细讨论了从故障传播模型中构造贝叶斯网络,以及分布式告警相关性模型的实现框架。最后结合SDH over DWDM系统,具体分析了基于贝叶斯网络的告警相关性分析过程及实验结果。实验证明利用贝叶斯网络能够准确定位通信网根故障点。     

随机petri网分析分组交换网中窗式流量控制机理

本文将随机Ｐｅｔｒｉ网与排队论相结合，对分组交换网中的窗式流量控制机理进行了描述与分析，建立了窗式流量控制机理的随机Ｐｅｔｒｉ网模型，并给出了随机Ｐｅｔｒｉ网模型的可达图及状态转移方程。为通信网的性能分析和评价提供了一种新的方法。     

一种通过节点序寻优进行贝叶斯网络结构学习的算法

针对K2算法过度依赖节点序,遗传算法节点序寻优效率差的问题,该文提出一种直接对节点序进行评分搜索的贝叶斯结构学习算法。该算法以K2算法为基础,首先通过计算支撑树权重矩阵,构建能够定量评价节点序的适应度函数。然后通过提出混合交叉策略和孤立节点处理机制,同时利用动态学习因子和倒置变异策略,提升遗传算法节点序寻优的性能。最后将得到的节点序作为K2算法的先验知识得到最优贝叶斯网络结构。仿真结果表明,该方法解决了K2算法依赖先验知识的问题,相比于其它优化算法,评分值平均增加了13.11%。     

贝叶斯网络在电子产品可靠性分析中的应用

讨论了传统可靠性分析方法的优点和缺点,简述了贝叶斯网络的优点及其因果推理与诊断推理,详细讨论了桶消元法的步骤。用一个算例说明贝叶斯网络的推理过程,结果表明,贝叶斯网络的双向推理可以有效地识别电子产品的薄弱环节,为进一步提高电子产品的可靠性及提高维修效率提供依据。