基于概率的剪枝算法

以alpha-beta剪枝算法为研究对象,提出一种基于alpha-beta剪枝和概率剪枝因素相结合的概率剪枝算法,来解决博弈树搜索问题。利用概率剪枝算法,可减少博弈树搜索深度,从而加快搜索进程。     

基于概率的剪枝算法

以alpha—beta剪枝算法为研究对象，提出一种基于alpha—beta剪枝和概率剪枝因素相结合的概率剪枝算法．来解决博弈树搜索问题。利用概率剪枝算法，可减少博弈树搜索深度，从而加快搜索进程。     

概率假设密度高斯混合实现的分量删减

针对概率假设密度(Probability hypothesis density, PHD)高斯混合实现算法中的分量删减问题, 提出了基于Dirichlet分布的分量删减算法以改进概率假设密度高斯混合实现算法的性能. 算法采用极大后验准则估计混合参数, 采用仅依赖于混合权重的负指数Dirichlet分布作为混合参数的先验分布, 利用拉格朗日乘子推导了混合权重的更新公式. 算法利用负指数Dirichlet分布的不稳定性,在极大后验迭代过程中驱使与目标强度不相关的分量消亡. 该不稳定性还能够解决多个相近分量共同描述一个强度峰值的问题, 有利于后续多目标状态的提取. 仿真结果表明, 基于Dirichlet分布的分量删减算法优于典型高斯混合实现中的删减算法.     

Increasing the effectiveness of associative classification in terms of class imbalance by using a novel pruning algorithm

Having received considerable interest in recent years, associative classification has focused on developing a class classifier, with lesser attention paid to the probability classifier used in direct marketing. While contributing to this integrated framework, this work attempts to increase the prediction accuracy of associative classification on class imbalance by adapting the scoring based on associations (SBA) algorithm. The SBA algorithm is modified by coupling it with the pruning strategy of association rules in the probabilistic classification based on associations (PCBA) algorithm, which is adjusted from the CBA for use in the structure of the probability classifier. PCBA is adjusted from CBA by increasing the confidence through under-sampling, setting different minimum supports (minsups) and minimum confidences (minconfs) for rules of different classes based on each distribution, and removing the pruning rules of the lowest error rate. Experimental results based on benchmark datasets and real-life application datasets indicate that the proposed method performs better than C5.0 and the original SBA do, and the number of rules required for scoring is significantly reduced.     

An algorithm to compute the minimal siphons in S 4 PR nets

Minimal siphons in the class of S ⁴ PR nets have become a conceptual and practical central tool for the study of the resource allocation related aspects in discrete event dynamic systems as, for example, the existence of deadlocks. Therefore the availability of efficient algorithms to compute the minimal siphons is essential. In this paper we try to take advantage of the particular properties of the siphons in S ⁴ PR to obtain an efficient algorithm. These properties allow us to express minimal siphons as the union of pruned minimal siphons containing only one resource. The pruning operation is built from the binary pruning relation defined on the set of minimal siphons containing only one resource. This pruning relation is represented by means of a directed graph. The computation of the minimal siphons is based on the maximal strongly connected components of this graph. The algorithm is highly economic in memory in all intermediate steps when compared to the classical algorithms.     

Fast 2-D 8×8 discrete cosine transform algorithm for image coding

A new fast two-dimension 8×8 discrete cosine transform (2D 8×8 DCT) algorithm based on the charac-teristics of the basic images of 2D DCT is presented. The new algorithm computes each DCT coefficient in turn more independently. Hence,the new algorithm is suitable for 2D DCT pruning algorithm of prun-ing away any number of high-frequency components of 2D DCT. The proposed pruning algorithm is more efficient than the existing pruning 2D DCT algorithms in terms of the number of arithmetic opera-tions,especially the number of multiplications required in the computation.     

多目标随机运输路径选择的频域优化模型

根据运输系统的随机特性,讨论时间、损耗和流量等优化目标之间的函数关系,采用概率论方法提出一种用于搜索时变、随机运输网络中多目标路径优化的频域生成图模型(FSG),设计相应的优化算法。FSG通过时频域间概率函数的相互转化,可定量分析O-D对之间多目标路径选择概率的动态变化过程,处理连续概率分布和离散经验分布。结合Matlab给出的算例验证了该算法的可行性和有效性。     

引入概率分布的深度神经网络贪婪剪枝

目的 深度学习在自动驾驶环境感知中的应用,将极大提升感知系统的精度和可靠性,但是现有的深度学习神经网络模型因其计算量和存储资源的需求难以部署在计算资源有限的自动驾驶嵌入式平台上。因此为解决应用深度神经网络所需的庞大计算量与嵌入式平台有限的计算能力之间的矛盾,提出了一种基于权重的概率分布的贪婪网络剪枝方法,旨在减少网络模型中的冗余连接,提高模型的计算效率。方法 引入权重的概率分布,在训练过程中记录权重参数中较小值出现的概率。在剪枝阶段,依据训练过程中统计的权重概率分布进行增量剪枝和网络修复,改善了目前仅以权重大小为依据的剪枝策略。结果 经实验验证,在Cifar10数据集上,在各个剪枝率下本文方法相比动态网络剪枝策略的准确率更高。在ImageNet数据集上,此方法在较小精度损失的情况下,有效地将AlexNet、VGG（visual geometry group）16的参数数量分别压缩了5.9倍和11.4倍,且所需的训练迭代次数相对于动态网络剪枝策略更少。另外对于残差类型网络ResNet34和ResNet50也可以进行有效的压缩,其中对于ResNet50网络,在精度损失增加较小的情况下,相比目前最优的方法HRank实现了更大的压缩率（2.1倍）。结论 基于概率分布的贪婪剪枝策略解决了深度神经网络剪枝的不确定性问题,进一步提高了模型压缩后网络的稳定性,在实现压缩网络模型参数数量的同时保证了模型的准确率。     

多目标随机运输路径选择的频域优化模型

根据运输系统的随机特性,讨论时间、损耗和流量等优化目标之间的函数关系,采用概率论方法提出一种用于搜索时变、随机运输网络中多目标路径优化的频域生成图模型(FSG),设计相应的优化算法。FSG通过时频域间概率函数的相互转化,可定量分析O-D对之间多目标路径选择概率的动态变化过程,处理连续概率分布和离散经验分布。结合Matlab给出的算例验证了该算法的可行性和有效性。     

Supporting early pruning in top-k query processing on massive data

This paper analyzes the execution behavior of “No Random Accesses” (NRA) and determines the depths to which each sorted file is scanned in growing phase and shrinking phase of NRA respectively. The analysis shows that NRA needs to maintain a large quantity of candidate tuples in growing phase on massive data. Based on the analysis, this paper proposes a novel top-k algorithm Top-K with Early Pruning (TKEP) which performs early pruning in growing phase. General rule and mathematical analysis for early pruning are presented in this paper. The theoretical analysis shows that early pruning can prune most of the candidate tuples. Although TKEP is an approximate method to obtain the top-k result, the probability for correctness is extremely high. Extensive experiments show that TKEP has a significant advantage over NRA.     

Depth-limited search for real-time problem solving

We propose depth-limited heuristic search as a general paradigm for real-time problem solving in a dynamic environment. When combined with iterative-deepening, it provides the ability to commit to an action almost instantaneously, but allows the quality of that decision to improve as long as time is available. Once a deadline is reached, the best decision arrived at is executed. We illustrate the paradigm in three different settings, corresponding to single-agent search, two-player games, and multi-agent problem solving. First we review two-player minimax search with alpha-beta pruning. Minimax can be extended to themaxn algorithm for more than two players, which admits a much weaker form of alpha-beta pruning. Finally, we explore real-time search algorithms for single-agent problems. Minimax is specialized tominimin, which allows a very powerfulalpha pruning algorithm. In addition,real-time-A ^* allows backtracking while still guaranteeing a solution and making locally optimal decisions.This research was supported by an NSF Presidential Young Investigator Award, NSF Grant IRI-8801939, and and equipment grant from Hewlett-Packard. Thanks to Valerie Aylett for drawing the figures.     

Minimax search algorithms with and without aspiration windows

Investigation of several algorithms for computing exact minimax values of game trees (utilizing backward pruning) are discussed. The focus is on trees with an ordering similar to that actually found in game playing practice. The authors compare the algorithms using two different distributions of the static values, the uniform distribution and a distribution estimated from practical data. A systematic comparison of using aspiration windows for all of the usual minimax algorithms is presented. The effects of aspiration windows of varying size and position are analyzed. Increasing the ordering of moves to near the optimum results in unexpectedly high savings. Algorithms with linear space complexity benefit most. Although the ordering of the first move is of predominant importance, that of the remainder has only second-order effects. The use of an aspiration window not only makes alpha-beta search competitive, but there also exist dependencies of its effects on certain properties of the trees     

Finite-dimensional recurrent algorithms for optimal nonlinear logical–dynamical filtering

The problem of the most accurate estimation of the current state of a multimode nonlinear dynamic observation system with discrete time based on indirect measurements of this state is considered. The general case when a mode indicator is available and the measurement errors depend on the plant disturbances is investigated. A comparative analysis of two known approaches is performed—the conventional absolutely optimal one based on the use of the posterior probability distribution, which requires the use of an unimplementable infinite-dimensional estimation algorithm, and a finitedimensional optimal approach, which produces the best structure of the difference equation of a low-order filter. More practical equations for the Gaussian approximations of these two optimal filters are obtained and compared. In the case of the absolutely optimal case, such an approximation is finitedimensional, but it differs from the approximation of the finite-dimensional optimal version in terms of its considerably larger dimension and the absence of parameters. The presence of parameters, which can be preliminarily calculated using the Monte-Carlo method, allows the Gaussian finite-dimensional optimal filter to produce more accurate estimates.     

A Heuristic Sampling Method for Maintaining the Probability Distribution

Sampling is a fundamental method for generating data subsets. As many data analysis methods are deve-loped based on probability distributions, maintaining distributions when sampling can help to ensure good data analysis performance. However, sampling a minimum subset while maintaining probability distributions is still a problem. In this paper, we decompose a joint probability distribution into a product of conditional probabilities based on Bayesian networks and use the chi-square test to formulate a sampling problem that requires that the sampled subset pass the distribution test to ensure the distribution. Furthermore, a heuristic sampling algorithm is proposed to generate the required subset by designing two scoring functions: one based on the chi-square test and the other based on likelihood functions. Experiments on four types of datasets with a size of 60000 show that when the significant difference level,α, is set to 0.05, the algorithm can exclude 99.9％, 99.0％, 93.1％ and 96.7％ of the samples based on their Bayesian networks—ASIA, ALARM, HEPAR2, and ANDES, respectively. When subsets of the same size are sampled, the subset generated by our algorithm passes all the distribution tests and the average distribution difference is approximately 0.03; by contrast, the subsets generated by random sampling pass only 83.8％of the tests, and the average distribution difference is approximately 0.24.     

基于熵分布的概率假设密度滤波器高斯混合实现

针对概率假设密度滤波器, 提出一种基于熵分布的高斯混合实现算法. 在该算法中, 作为混合参数先验分布的熵分布, 主要用在极大后验迭代过程中删减无关混合分量, 该删减操作可通过混合权重调整来实现. 此外, 该算法还能够解决多个具有类似参数的混合分量共同描述一个强度峰值的问题. 仿真结果表明, 所提出算法优于典型的阈值删减算法.     

博弈树启发式搜索的α-β剪枝技术研究

博弈是启发式搜索的一个重要应用领域,博弈的过程可以用一棵博弈搜索树表示,通过对博弈树进行搜索求取问题的解,搜索策略常采用α-β剪枝技术。在深入研究α-β剪枝技术的基础上,提出在扩展未达到规定深度节点时,对扩展出的子节点按照估价函数大小顺序插入到搜索树中,从而在α-β剪枝过程中剪掉更多的分枝,提高搜索效率。     

不确定数据频繁闭项集挖掘算法

由于不确定数据的向下封闭属性,挖掘全部频繁项集的方法会得到一个指数级的结果。为获得一个较小的合适的结果集,研究了在不确定数据上挖掘频繁闭项集,并提出了一种新的频繁闭项集挖掘算法——NA-PFCIM。该算法将项集挖掘过程看作一个概率分布函数,考虑到基于正态分布模型的方法提取的频繁项集精确度较高,而且支持大型数据库,采用了正态分布模型提取频繁项集。同时,为了减少搜索空间以及避免冗余计算,利用基于深度优先搜索的策略来获得所有的概率频繁闭项集。该算法还设计了两个剪枝策略:超集修剪和子集修剪。最后,在常用的数据集(T10I4D100K、Accidents、Mushroom、Chess)上,将提出的NA-PFCIM算法和基于泊松分布的A-PFCIM算法进行比较。实验结果表明,NA-PFCIM算法能够减少所要扩展的项集,同时减少项集频繁概率的计算,其性能优于对比算法。     

A variable iterated greedy algorithm with differential evolution for the no-idle permutation flowshop scheduling problem

This paper presents a variable iterated greedy algorithm (IG) with differential evolution (vIG_DE), designed to solve the no-idle permutation flowshop scheduling problem. In an IG algorithm, size d of jobs are removed from a sequence and re-inserted into all possible positions of the remaining sequences of jobs, which affects the performance of the algorithm. The basic concept behind the proposed vIG_DE algorithm is to employ differential evolution (DE) to determine two important parameters for the IG algorithm, which are the destruction size and the probability of applying the IG algorithm to an individual. While DE optimizes the destruction size and the probability on a continuous domain by using DE mutation and crossover operators, these two parameters are used to generate a trial individual by directly applying the IG algorithm to each target individual depending on the probability. Next, the trial individual is replaced with the corresponding target individual if it is better in terms of fitness. A unique multi-vector chromosome representation is presented in such a way that the first vector represents the destruction size and the probability, which is a DE vector, whereas the second vector simply consists of a job permutation assigned to each individual in the target population. Furthermore, the traditional IG and a variable IG from the literature are re-implemented as well. The proposed algorithms are applied to the no-idle permutation flowshop scheduling (NIPFS) problem with the makespan and total flowtime criteria. The performances of the proposed algorithms are tested on the Ruben Ruiz benchmark suite and compared to the best-known solutions available at http://soa.iti.es/rruiz as well as to those from a recent discrete differential evolution algorithm (HDDE) from the literature. The computational results show that all three IG variants represent state-of-art methods for the NIPFS problem.     

基于高斯概率模型的字符识别算法的研究

提出了一种新的基于高斯概率模型的字符识别算法,该算法根据模式识别的样本分布特征与高斯分布的一致性,构建了一个高斯概率模型.在模型中存储概率为P的训练样本,分类识别时,将测试样本与模型进行相关计算得出概率值,进行判断.结果表明,该算法识别速度快,准确率高,与其他字符识别算法(KNN)相比有更好的实用性.     

基于强化学习的无标签网络剪枝

为了消除深度神经网络中的冗余结构,找到具备较好性能和复杂度之间平衡性的网络结构,提出基于无标签的全局学习方法(LFGCL).LFGCL学习基于网络体系结构表示的全局剪枝策略,可有效避免以逐层方式修剪网络而导致的次优压缩率.在剪枝过程中不依赖数据标签,输出与基线网络相似的特征,优化网络体系结构.通过强化学习推断所有层的压...