算法

Based on the study of Likelihood Ratio Test voice activity detection algorithm and discussion about decision-directed LRT and smoothed LRT voice activity detection algorithms, we proposed a novel voice activity detection algorithm for improving speech detection robustness in noisy environments.

深入研究了基于似然比测试的语音激活检测算法，并对基于直接决策似然比测试的语音激活检测算法和基于平滑似然比测试的语音激活检测算法进行了讨论。

In the end.we prove that the limit probability distribution of the Markov chains exists.

特别讨论了一种基于特殊选择的抽象进化算法，根据状态空间的分类，分析了这种特殊选择进化算法种群的一步转移概率矩阵的性质，最后证明了这种算法的Markov链存在极限概率分布。

A new method of slope stability, limit equilibrium finite element method based on Dijkstra algorithm, was established by applying the improved Dijkstra algorithm which was an algorithm to seek the shortest path in graph theory to search for the minimum safety factor of slopes and the position of most dangerous sliding plane.

通过引入图论中解决最短路问题的Dijkstra算法到搜索边坡最危险滑动面及其安全系数中去，对Dijkstra算法作了一定的改进，建立了一种新的边坡稳定性分析方法―基于Dijkstra算法的极限平衡有限元方法。

Fast Line Integral Convolution with constant filter kernels as well as with triangular filter kernels are presented,which are faster than the original Line Integral Convolution approach about an order of magnitude in rendering actual vector fields.

针对这个问题实现了基于常数滤波卷积核的快速线积分卷积算法，提出并论证了基于三角形滤波卷积核的快速线积分卷积算法，对于实际矢量场，上述两种方法均将运算速度提高了一个数量级，而基于三角形滤波卷积核的快速线积分卷积算法的结果使图像更加平滑。

The forth algorithm is the same as the third one with an addition of finite termination criteria. Although the latter two algorithms have only linear rate of convergence, they are especially suitable for large-scale and sparse problems, with features of simple formula, small storage, sparsity preservation and easy implementation.

前者借助NCP函数把互补问题转化为等价的非光滑方程组，再用带参数的光滑方程组近似这些非光滑方程组，最后用牛顿型方法求解所得到的光滑方程组，希望通过光滑参数趋于零得到原来互补问题的解；后者基于等价不动点格式，构造了一个光滑迭代算法和一个具有有限终止性质的算法，虽然这种迭代算法仅有线性收敛速度，但由于其格式简单、存储量小、保稀疏性、非常易于计算机实现等特点，故较适用于求解大规模稀疏问题。

LPP has incorporated the advantages of both linear and nonlinear dimension reduction algorithms.

LPP算法是基于LE算法的线性扩展，兼具有非线性的LE算法和线性降维方法二者的优点。

The Locally Linear Embedding algorithm based on Manifold learning is introduced firstly in this paper,because the

从分析基于流形学习理论的局部线性嵌入算法入手，针对传统的局部线性嵌入算法在源数据稀疏时会失效的缺点，提出了基于局部线性逼近思想的流形学习算法，并在S-曲线上采样测试取得良好降维效果。

The data in LLE are optimized using the small world algorithm, and the shortest path and the local neighbor set clustering coefficients are used as the local parameters. As a result, the problem of the embedding distortion using only local linear patch of the manifold to define neighborhood in Euclidean space is effectively solved. Three groups of standard data sets are selected to test and to compare the efficiency and robustness of SLLE and LLE.

将复杂网络算法引入到流形学习中，利用小世界算法对LLE算法进行数据优化，并以最短路径和局部集群系数作为局部优化参数，解决了数据点不规则时以欧氏空间作为邻域判别标准在构建局部超平面造成嵌入结果扭曲的难题。

As to the multiple classification algorithms of large size rule set, we introduce a algorithms named recursive flow classification and this algorithms use the linear buffer to implement recursive map.

接着本文介绍了应用于大规模规则库的快速分类算法的解决方案——递归流分类，该算法是一种利用线性存储区分块递归映射的算法。

According to the linear search technique, the Quasi-Newton algorithm can be divided into monotone algorithm and non-monotone algorithm.

根据所采用的搜索准则，拟牛顿算法可以分为单调的算法和非单调的算法。

The split between the two groups can hardly be papered over.

这两个团体间的分歧难以掩饰。

This approach not only encourages a greater number of responses, but minimizes the likelihood of stale groupthink.

这种做法不仅鼓励了更多的反应，而且减少跟风的可能性。