發(fā)布時(shí)間：2018-05-26 12:05

  本文選題：復(fù)雜網(wǎng)絡(luò) + 社團(tuán)發(fā)現(xiàn)��； 參考：《計(jì)算機(jī)應(yīng)用》2017年11期

【摘要】：經(jīng)典的置信傳播(BP)算法能夠通過(guò)有限次數(shù)的迭代,推斷出所有節(jié)點(diǎn)的邊緣概率分布和最大似然概率。針對(duì)該算法在迭代過(guò)程中產(chǎn)生的影響精度和收斂速度的強(qiáng)烈震蕩,找出了造成震蕩的三個(gè)主要因素:強(qiáng)勢(shì)能、緊密的環(huán)路和矛盾的方向,并有針對(duì)性地改進(jìn)了該算法的核心更新規(guī)則;同時(shí)又進(jìn)一步提出了異步消息傳遞方式,克服傳統(tǒng)置信傳播算法采用的同步消息傳播方式的收斂慢、效率低等缺點(diǎn)。利用隨機(jī)塊模型擬合網(wǎng)絡(luò)的生成過(guò)程,利用經(jīng)典的期望最大化算法對(duì)模型進(jìn)行求解,分別利用改進(jìn)前后的置信傳播算法推斷隱變量的后驗(yàn)概率。在五個(gè)真實(shí)網(wǎng)絡(luò)上的實(shí)驗(yàn)表明,兩個(gè)改進(jìn)均使得精度和速度不同程度地提高。
[Abstract]:The classical confidence Propagation (BP) algorithm can infer the edge probability distribution and the maximum likelihood probability of all nodes through a finite number of iterations. Aiming at the strong oscillation of the accuracy and convergence rate of the algorithm in the iterative process, three main factors are found out: strong energy, tight loop and contradictory direction. The core updating rules of the algorithm are improved and the asynchronous message passing method is proposed to overcome the shortcomings of slow convergence and low efficiency of the synchronous message transmission method used in the traditional confidence propagation algorithm. The stochastic block model is used to fit the generating process of the network, the classical expectation maximization algorithm is used to solve the model, and the improved confidence propagation algorithm is used to infer the posterior probability of the hidden variables. Experiments on five real networks show that both the two improvements improve the accuracy and speed to varying degrees.
【作者單位】： 天津大學(xué)軟件學(xué)院;
【基金】：國(guó)家自然科學(xué)基金資助項(xiàng)目(61303110,61502334) 天津大學(xué)北洋學(xué)者·青年骨干教師項(xiàng)目(2017XRG-0016)~~
【分類(lèi)號(hào)】：O157.5;TP301.6
，

本文編號(hào)：1937173