AN ITERATIVE INFERENCE MECHANISM FOR THE PROBABILISTIC EXPERT SYSTEM PES

Abstract

The paper suggests a new inference mechanism based on iterative use of the Bayesian inference scheme. The procedure iteratively computes optimal component weights of a distribution mixture from a class called generalized knowledge base. It is proved that the iterative process converges to a unique limit whereby the resulting probability distribution can be defined as the information-divergence projection of input distribution on the generalized knowledge base. The iterative inference mechanism resembles natural process of cognition as iteratively improving understanding of the input information.     

归纳学习算法CAP2的研究与应用

本文提出以实例空间中状态划分概率的大小作为启发式信息，以提供的正反实例集为依据，基于二叉树分类方法的示例式归纳学习算法ＣＡＰ２．它输出的分类规则是谓词演算表达式．该算法可根据用户对精度的要求控制分类深度，得到不同精度的规则，并能处理连续数据、噪音数据和利用用户提供的背景知识，既适用于同时给定概念的正、反例集的情况，也适用于只给正例集的情况．本文还介绍了ＣＡＰ２算法的应用情况，并和著名的ＩＤ３算法进行了比较．ＣＡＰ２已嵌入到一个自动知识获取系统．     

AN ENHANCED ANT COLONY OPTIMIZATION METAHEURISTIC FOR THE MINIMUM DOMINATING SET PROBLEM

This paper proposes an enhanced Ant Colony Optimization (ACO) metaheuristic called ACO-TS to attack the minimum dominating set (MDS) problem. One of the recognized difficulties faced by ACO in its original form is premature convergence, which produces less satisfactory solutions. We propose a way to encourage a higher degree of exploration of the search space by incorporating a technique based on a concept borrowed from genetic algorithms called tournament selection. Instead of always following the standard mechanism for selecting the next solution component, an ant would make its decision based on the outcome of a tournament between randomly selected allowable components. The frequency of the tournament selection is controlled by a probability measure. The use of tournament selection is coupled with an iteration-best pheromone update. To evaluate the enhanced ACO, we consider the MDS problem formulated from ad hoc network clustering. A comparison with its original form shows that the enhanced ACO produces better solutions using fewer number of cycles. We also empirically demonstrate that the proposed ACO produces better solutions than a genetic algorithm. Finally, we argue, based on empirical results, why the tournament selection approach is preferable to a pure random selection method.     

基于图卷积网络的知识感知对话生成

引入结构化知识的对话系统因为能够生成流畅度更高、多样性更丰富的对话回复而受到广泛关注, 但是以往的研究只注重于结构化知识中的实体, 却忽略了实体之间的关系以及知识的完整性. 本文提出了一种基于图卷积网络的知识感知对话生成模型(KCG). 该模型通过知识编码器分别捕获实体与关系的语义信息并利用图卷积网络增强实体表征; 再利用知识选择模块获得与对话上下文相关的实体与关系的知识选择概率分布; 最后将知识选择概率分布与词表概率分布融合, 解码器以此选择知识或词表字词. 本文在中文公开数据集DuConv上进行实验, 结果表明, KCG在自动评估指标上优于目前的基线模型, 能生成更加流畅并且内容更加丰富的回复.     

基于半监督聚类的人脸检测方法

将肤色与连续AdaBoost算法相结合进行人脸检测,并引入半监督策略指导肤色聚类从而建立肤色模型。在肤色聚类过程中,提出一种基于半监督的SKDK算法引导肤色聚类,依据各个像素簇的概率统计分布特性得到肤色模型。在此基础上利用数学形态学等知识对图像进行处理,得到人脸候选区域,将其作为连续AdaBoost分类器的输入进行人脸检测。实验结果表明,在多人脸的场景下,该方法的检测效果优于直接使用连续AdaBoost方法进行人脸检测的检测效果。     

基于分布估计算法的连续函数全局优化问题研究

分布估计算法从宏观的角度建立一个概率模型,用来描述解空间的分布,从而通过进化计算获得优势个体。目前,离散型分布估计算法研究已经比较成熟,而连续型分布估计算法研究进展缓慢。采用均匀分布缩小采样领域的思想,设计新的分布估计算法求解连续型优化问题。实验数据表明,该分布估计算法对于求解连续型问题是有效的。     

Generalized Jeffrey’s rule of conditioning and evidence combining rule for a priori probabilistic knowledge in conditional evidence theory

In Bayesian probabilistic approach for uncertain reasoning, one basic assumption is that a priori knowledge about the uncertain variable is modeled by a probability distribution. When new evidence representable by a constant set is available, the Bayesian conditioning is used to update a priori knowledge. In the conventional D-S evidence theory, all bodies of evidence about the uncertain variable are imprecise and uncertain. All bodies of evidence are combined by so-called Dempster’s rule of combination to achieve a combined body of evidence without considering a priori knowledge. From our point of view, when identifying the true value of an uncertain variable, Bayesian approach and evidence theory can cooperate to deal with uncertain reasoning. Firstly all imprecise and uncertain bodies of evidence about the uncertain variable are fused to achieve a combined evidence based on a priori knowledge, then the a posteriori probability distribution is achieved from a priori probability distribution by conditioning on the combined evidence. In this paper we firstly deal with the knowledge updating problem where a priori knowledge is represented by a probability distribution and new evidence is represented by a random set. Then we review the conditional evidence theory which resolves the knowledge combining problem based on a priori probabilistic knowledge. Finally we discuss the close relationship between knowledge updating procedure and knowledge combining procedure presented in this paper. We show that a posteriori probability conditioned on fused body of evidence satisfies the Bayesian parallel combination rule.     

图象重建的最小距离算法

本文提出了图象重建的一种凸集投影算法.它的重建图象是所有满足投影约束的图象中与先验图象的距离最小者.该算法是就连续分布的图象导出的,重建图象时也不需在空域与频域间进行变换,是一种较OSPR方法更直接、更简单的方法.还对最小距离算法重建结果的存在性、唯一性、幂等性等性质作了证明.     

关于最小分散镇定结构的研究

本文考察线性时不变多变量系统的分散镇定问题,揭示了局部控制站间的通信与消除固定模间的内在联系,并由此把求最小(最经济)分散可镇定结构问题转化成一个显式的特殊0-1规划问题,导出了一种求最小分散可镇定结构的有效算法。     

基于概率模型和倒谱差分的特征补偿算法

在概率模型中，给出了引入倒谱预测值的动态相关性来进行特征补偿的方法。该方法采用期望最大化(EM)算法来估计联合分布参数，基于语音和噪声的先验概率密度，在倒谱域中对语音特征参数进行最小均方误差预测(MMSE)，以提高语音识别精度。不同噪声环境和不同信噪比下的实验结果表明，该方法能有效地提高噪声环境下的中文连续语音识别的正确率。     

融合概率分布和单调性的支持向量回归算法

传统支持向量回归是单纯基于样本数据的输入输出值建模,仅使用样本数据信息,未充分利用其他已知信息,模型泛化能力不强.为了进一步提高其性能,提出一种融合概率分布和单调性先验知识的支持向量回归算法.首先将对偶二次规划问题简化为线性规划问题,在求解时,加入与拉格朗日乘子相关的单调性约束条件;通过粒子群算法优化惩罚参数和核参数,优化目标包括四阶矩估计表示的输出样本概率分布特性.实验结果表明,融合这两部分信息的模型,能使预测值较好地满足训练样本隐含的概率分布特性及已知的单调性,既提高了预测精度,又增加了模型的可解释性.     

Efficient and robust probabilistic guarantees for real-time tasks

This paper presents a new method for providing probabilistic real-time guarantees to tasks scheduled through resource reservations. Previous work on probabilistic analysis of reservation-based schedulers is extended by improving the efficiency and robustness of the probability computation. Robustness is improved by accounting for a possibly incomplete knowledge of the distribution of the computation times (which is typical in realistic applications). The proposed approach computes a conservative bound for the probability of missing deadlines, based on the knowledge of the probability distributions of the execution times and of the inter-arrival times of the tasks. In this paper, such a bound is computed in realistic situations, comparing it with simulative results and with the exact computation of deadline miss probabilities (without pessimistic bounds). Finally, the impact of the incomplete knowledge of the execution times distribution is evaluated.     

MEBN: A language for first-order Bayesian knowledge bases

Although classical first-order logic is the de facto standard logical foundation for artificial intelligence, the lack of a built-in, semantically grounded capability for reasoning under uncertainty renders it inadequate for many important classes of problems. Probability is the best-understood and most widely applied formalism for computational scientific reasoning under uncertainty. Increasingly expressive languages are emerging for which the fundamental logical basis is probability. This paper presents Multi-Entity Bayesian Networks (MEBN), a first-order language for specifying probabilistic knowledge bases as parameterized fragments of Bayesian networks. MEBN fragments (MFrags) can be instantiated and combined to form arbitrarily complex graphical probability models. An MFrag represents probabilistic relationships among a conceptually meaningful group of uncertain hypotheses. Thus, MEBN facilitates representation of knowledge at a natural level of granularity. The semantics of MEBN assigns a probability distribution over interpretations of an associated classical first-order theory on a finite or countably infinite domain. Bayesian inference provides both a proof theory for combining prior knowledge with observations, and a learning theory for refining a representation as evidence accrues. A proof is given that MEBN can represent a probability distribution on interpretations of any finitely axiomatizable first-order theory.     

Cpo's of measures for nondeterminism

In this approach to the semantics of nondeterminism, we introduce and study the complete partial order (cpo) of probability distributions on a domain. The approach avoids considering equivalent subsets, used in theory of powerdomains, which may lead to some unwelcome identifications. These results show that the class of probability distributions on a cpo is itself a cpo and that every probability distribution is the lub of an increasing sequence of ‘finite’ probability distributions. We introduce the probabilistic extensions of continuous functions in order to extend the ‘usual’ continuous functions on this new domain. On the other hand the structure of this cpo suggests introducing an operation called ‘random selection’, which is the counterpart of the ‘OR’ (or union) operation, commonly used in nondeterministic programs. The paper then studies the ‘naturalness’ of these extended notions and treats the question of continuity, which is of prime importance in the Scott theory of fixed point semantics.     

Imprecise prior knowledge incorporating into one-class classification

An extension of Campbell and Bennett’s novelty detection or one-class classification model incorporating prior knowledge is studied in the paper. The proposed extension relaxes the strong assumption of the empirical probability distribution over elements of a training set and deals with a set of probability distributions produced by prior knowledge about training data. The classification problem is solved by considering extreme points of the probability distribution set or by means of the conjugate duality technique. Special cases of prior knowledge are considered in detail, including the imprecise linear-vacuous mixture model and interval-valued moments of feature values. Numerical experiments show that the proposed models outperform Campbell and Bennett’s model for many real and synthetic data.     

The theory and practice of Bayesian image labeling

Image analysis that produces an image-like array of symbolic or numerical elements (such as edge finding or depth map reconstruction) can be formulated as a labeling problem in which each element is to be assigned a label from a discrete or continuous label set. This formulation lends itself to algorithms, based on Bayesian probability theory, that support the combination of disparate sources of information, including prior knowledge.In the approach described here, local visual observations for each entity to be labeled (e.g., edge sites, pixels, elements in a depth array) yield label likelihoods. Likelihoods from several sources are combined consistently in abstraction-hierarchical label structures using a new, computationally simple procedure. The resulting label likelihoods are combined with a priori spatial knowledge encoded in a Markov random field (MRF). From the prior probabilities and the evidence-dependent combined likelihoods, the a posteriori distribution of the labelings is derived using Bayes' theorem.A new inference method, Highest Confidence First (HCF) estimation, is used to infer a unique labeling from the a posteriori distribution that is consistent with both prior knowledge and evidence. HCF compares favorably to previous techniques, all equivalent to some form of energy minimization or optimization, in finding a good MRF labeling. HCF is computationally efficient and predictable and produces better answers (lower energies) while exhibiting graceful degradation under noise and least commitment under inaccurate models. The technique generalizes to higher-level vision problems and other domains, and is demonstrated on problems of intensity edge detection and surface depth reconstruction.     

THE CANONICAL STRUCTURE AS A MINIMUM STRUCTURE

In this paper a structure of a system is defined as a mathematical structure (?; Σ ), where ? is a first-order logic language and Σ is a set of sentences of the given first-order logic. It is shown that a canonical structure determined by Σ which is similar to those used in proving the Gödei's completeness theorem, satisfies a universality in the sense of category theory when homomorphisms are used as morphisms, and a freeness in the sense of universal algebra when Σ-morphisms, which preserve Σ are used. The universality and the freeness give the minimality of the canonical structure.

As an example, a structure of a stationary system is defined as a pair (?e-sta, Σe-sta)-Its canonical structure is actually constructed. In a sense this canonical structure accords with models constructed by Nerode realization.     

A PRINCIPLE OF UNCERTAINTY AND INFORMATION INVARIANCE*

The paper introduces a new principle, referred to as the principle of uncertainty and information invariance, for making transformations between different mathematical theories by which situations under uncertainty can be characterized. This principle requires that the amount of uncertainty (and related information) be preserved under these transformations. The principle is developed in sufficient details for transformations between probability theory and possibility theory under interval, log-interval and ordinal scales. Its broader use is discussed only in general terms and illustrated by an example.     

Initial approaches to the application of islands-based parallel EDAs in continuous domains

Estimation of distribution algorithms (EDAs) are a wide-ranging family of evolutionary algorithms whose common feature is the way they evolve by learning a probability distribution from the best individuals in a population and sampling it to generate the next one. Although they have been widely applied to solve combinatorial optimization problems, there are also extensions that work with continuous variables. In this paper [this paper is an extended version of delaOssa et al. Initial approaches to the application of islands-based parellel EDAs in continuous domains, in: Proceedings of the 34th International Conference on Parallel Processing Workshops (ICPP 2005 Workshops), Oslo, 2005, pp. 580–587] we focus on the solution of the latter by means of island models. Besides evaluating the performance of traditional island models when applied to EDAs, our main goal consists in achieving some insight about the behavior and benefits of the migration of probability models that this framework allow.