翻译规则剪枝与基于半强制解码和变分贝叶斯推理的模型训练

统计机器翻译一般采用启发式方法训练翻译模型。但启发式方法的理论基础不够完善,因此,会导致翻译模型规模庞大以及模型参数精确率不高。针对以上两个问题,该文提出一种基于变分贝叶斯推理的模型训练方法,形成更精确的精简翻译模型。该方法首先通过强制解码对齐语料,然后利用变分贝叶斯EM算法获得模型参数。该文的实验语料为NIST汉英翻译任务数据,实验结果显示,基于句法(基于短语)的统计机器翻译中,超过95%(76%)的规则被剪枝,且BLEU值显著提高。     

数据缺失条件下的贝叶斯推断方法

Recently Bayesian network(BN) becomus a noticeable research direction in Data Mining.In this paper we introduce missing data mechanisms firstly,and then some methods to do Baysesian inference with missing data based on these missing data mechanisms.All of these must be useful in practice especially when data is scare and expensive.It can foresee that Bayesian networks will become a powerful tool in Data Mining with all of these methods above offered.     

Interactive Instruction in Bayesian Inference

An instructional approach is presented to improve human performance in solving Bayesian inference problems. Starting from the original text of the classic Mammography Problem, the textual expression is modified and visualizations are added according to Mayer’s principles of instruction. These principles concern coherence, personalization, signaling, segmenting, multimedia, spatial contiguity, and pretraining. Principles of self-explanation and interactivity are also applied. Four experiments on the Mammography Problem showed that these principles help participants answer the questions at significantly improved rates. Nonetheless, in novel interactivity conditions, performance was lowered suggesting that more interaction can add more difficulty for participants. Overall, a leap forward in accuracy was found, with more than twice the participant accuracy of previous work. This indicates that an instructional approach to improving human performance in Bayesian inference is a promising direction.     

贝叶斯网络推理的研究与分析

贝叶斯网络是人工智能中不确定知识表示和推理的有力工具.介绍了贝叶斯网络的概念,给出一个实例,分析了贝叶斯网络推理的方法和过程.     

贝叶斯网络推理的研究与分析

贝叶斯网络是人工智能中不确定知识表示和推理的有力工具。介绍了贝叶斯网络的概念,给出一个实例,分析了贝叶斯网络推理的方法和过程。     

基于贝叶斯推理的目标跟踪

该文描述了基于贝叶斯推理的目标跟踪算法,可应用于非线性、非高斯系统中。介绍了贝叶斯跟踪基本概念。讨论了单目标跟踪算法及在不考虑数据关联、考虑数据关联两种情况下的多目标跟踪算法。给出了每种跟踪算法应满足的假设条件及其递归方程。最后,简要介绍了在工程中应用的具体算法。     

Cognitive Architecture,Holistic Inference and Bayesian Networks

Minds and Machines - Two long-standing arguments in cognitive science invoke the assumption that holistic inference is computationally infeasible. The first is Fodor’s skeptical argument...     

热电偶动态响应模型的贝叶斯推断

热电偶对温度激励的动态响应受到测量接点的体积变化、比热容变化等多种因素的影响.为了评价热电偶的动态响应能力,采用气流环境下的阶跃法对热电偶产生温度激励,采集不同温度激励下热电偶的输出数据,得到热电偶的动态响应数据集.采用时间序列分析方法建立了热电偶动态响应的回归模型,用贝叶斯方法进行模型的参数推断.采用马尔科夫链蒙特卡罗法中的杂交蒙特卡罗法抽样,计算参数的后验分布.为了提高抽样效率,使用最大后验估计值为抽样起始点并采用自适应方法调整步长.以一种镍铬镍硅热电偶为评价对象,给出了一个算例.     

基于文档图像的贝叶斯重建算法仿真研究

研究文档图像的分辨率提高问题,针对数字化文档图像在采集过程中遇到的低分辨率、噪声、纸张质量蜕化和形变等因素影响,提出了一种新的贝叶斯估计的最大后验概率算法对文档图像进行恢复和重构.首先利用聚类方法对文档中文字进行自动分类,然后依据每个类别中相同字符的先验知识,例如出现频率,几何特性等,利用一个能量方程来求取最终的MAP最优解,然后一个新颖的MAP迭代算法,反复利用对高分辨率图像的估计来逼近最优解,从而使得最终的高分辨率字符图像获得很高的清晰度.仿真结果表明提出的算法能稳定地提高文字的分辨率,提高文档的识别准确率,并且具有高的运算效率.在此基础上利用本文方法,可以方便的实现多文档或者书籍图像的重建和恢复.     

基于全局消息传播的贝叶斯推理

在贝叶斯网络中，常常需要作不确定概率推理。然而针对一般复杂网络，精确推理算法由于计算复杂度太高而常常被摒弃。针对这一问题，本文提出了一种基于全局传播的PPJT近似推理算法。PPJT算法采用消息传播机制，通过消息的收集与分发过程，可以更新和修正连接树节点的团势并最终生成相容连接树。与另一种常用的近似推理算法即似然权重（Likelihood Weighting）算法的时间性能对比实验显示，采用消息传播机制的PPJT算法有效地降低了计算的时间复杂度；同时与似然权重算法的性能对比实验表明，在相对小规模观察样本输入条件下，PPJT算法能够保证更高的概率推理精度。PPJT算法为实现一般复杂网络中的概率推理提供了一种新的理论工具。     

Using the BACC Software for Bayesian Inference

The BACC software provides its users with tools for Bayesian Analysis,Computation and Communications. The current version of the software, describedhere, implements these tools as extensions to popular mathematicalapplications such as Matlab, S-PLUS, R, and Gauss, running under Windows,Linux or Unix. From the user's perspective, there is a seamless integrationof special-purpose BACC commands for posterior simulation and related taskswith powerful built-in commands for matrix computation, graphics, program flowcontrol, and I/O. Examples demonstrate the use of the software within Matlab.Nineteen models are currently available, and many others are planned. BACC isdesigned to be extendible, not only by the developers of BACC, but also byothers who wish to implement their own models and thus make them available toBACC users. While model development requires programming in C, several designfeatures of BACC facilitate this development. BACC is freely available athttp://www2.cirano.qc.ca/bacc/.     

基于三维特征参数的贝叶斯推理电路功耗模型

在功耗与信号统计分析的基础上,采用贝叶斯推理技术建立周期精确的功耗宏模型.通过分析信号特征对电路功耗的影响,选择输入信号密度Pin、输入跳变密度Din和输出跳变密度Dout作为贝叶斯推理的三维特征参数,证明了上述特征参数对信号时间和空间相关性信息的覆盖.实验结果表明,该方法较目前的门级功耗分析速度提高400余倍,周期功耗平均误差可以控制在10%以内.     

自适应推导下的统一化调试加速技术

在传统调试过程中,缺陷定位通常作为程序修复的前置步骤.最近,一种新型调试框架(统一化调试)被提出.不同于传统调试中缺陷定位和程序修复的单向连接方式,统一化调试首次建立了定位与修复之间的双向连接机制,从而达到同时提升两个领域的效果.作为首个统一化调试技术,ProFL利用程序修复过程中伴随产生的大量补丁执行信息逆向地提升已...     

Convolutional Networks with Adaptive Inference Graphs

International Journal of Computer Vision - Do convolutional networks really need a fixed feed-forward structure? What if, after identifying the high-level concept of an image, a network could move...     

基于贝叶斯推理与PFNET理论的认知模型

针对从固定认知结构中生成认知模型的局限性,提出在认知元素固定而认知结构不固定的学习中使用贝叶斯推理方法和PFNET理论从以往学习者的样本信息中按学习者的要求生成"最佳Ki结点集合"和"最优Ki认知链",获得B-P认知模型;对B-P认知模型的生成原理进行说明并通过实例验证该模型的有效性和可行性。     

Inference Algorithms in Bayesian Networks and the Probanet System

This paper reviews and formalizes algorithms for probabilistic inferences uponcausal probabilistic networks(CPN), also known asBayesian networks,and introduces Probanet—a development environment for CPNs. Information fusion in CPNs is realized through updating joint probabilities of the variables upon the arrival of new evidences or new hypotheses. Kernel algorithms for some dominant methods of inferences are formalized from discontiguous, mathematics-oriented literatures, with gaps filled in with regards to computability and completeness. Probanet has been designed and developed as a generic shell, a development environment for CPN construction and application. The design aspects and current status of Probanet are described.     

贝叶斯网模型的学习、推理和应用

近年来在人工智能领域,不确定性问题一直成为人们关注和研究的焦点。贝叶斯网是用来表示不确定变量集合联合概率分布的图形模式,它反映了变量间潜在的依赖关系。使用贝叶斯网建模已成为解决许多不确定性问题的强有力工具。基于国内外最新的研究成果对贝叶斯网模型的学习、推理和应用情况进行了综述,并对未来的发展方向进行了展望。     

贝叶斯推断在MCDB分布式平台上的实现

提出了应用贝叶斯统计方法在分布式数据库MCDB上处理超大规模数据的实现方法,并以贝叶斯线性回归、话题模型的LDA和狄利克雷过程的聚类算法为例进行了论证.用户可以通过SQL语言定义变量之间的关系进行模拟.探索了一种使用简洁的SQL设计大规模统计学习系统的方法,其利用MCDB能够自动解决并行化和资源优化问题,以获得高性能的并行处理能力.     

Mathematical Fundamentals of Constructing Fuzzy Bayesian Inference Techniques

Problems and an associated technique for developing a Bayesian approach to decision-making in the case of fuzzy data are presented. The concept of fuzzy and pseudofuzzy quantities is introduced and main operations with pseudofuzzy quantities are considered. The basic relationships and the principal concepts of the Bayesian decision procedure based on the modus-ponens rule are proposed. Some problems concerned with the practical realization of the fuzzy Bayesian method are considered.     

一种用于信度网推理的高效三角化算法

信度网是不确定性知识表达和推理的有力工具。信度网的精确推理是NPC问题,计算的主要困难在于将信度网三角化并构造一棵最小权的join tree上。此项研究提出了一种新的三角化算法MsLB-Triang,该算法同时利用了无向图三角化的Direc性质与LB-单纯性质,在生成的三角化图的总权以及增加边的数目上均明显优于目前广泛采用的Min.Weight Heuristic算法。