针对局部线性嵌入算法(Local Linear Embedding,LLE)短路、离群点影响大和结构信息缺乏等问题,提出基于Kernel Rank-order距离的重构权重局部线性嵌入算法(Reconstruction weight Local Linear Embedding algorithm based on Kernel Ran...针对局部线性嵌入算法(Local Linear Embedding,LLE)短路、离群点影响大和结构信息缺乏等问题,提出基于Kernel Rank-order距离的重构权重局部线性嵌入算法(Reconstruction weight Local Linear Embedding algorithm based on Kernel Rank-order distance,KRLLE)。用核函数将样本点映射到高维使其更加线性可分,进而获得较好的近邻点集;计算重构权重系数进而得到加权重构权重,重构权重系数根据两点间相关性越大对重构贡献越大的特性来减小离群点的影响,并利用两点间的欧氏距离与测地线距离之比有效地将短路点排除在外;根据加权重构权重得到低维嵌入坐标。在ORL、Yale人脸库和MNIST手写体数据库上的实验表明,KRLLE对离群点具有更好的鲁棒性并且由于增加了结构信息,识别率得到了提高。展开更多
An improved parallel weighted bit-flipping（PWBF） algorithm is presented. To accelerate the information exchanges between check nodes and variable nodes, the bit-flipping step and the check node updating step of the ...An improved parallel weighted bit-flipping（PWBF） algorithm is presented. To accelerate the information exchanges between check nodes and variable nodes, the bit-flipping step and the check node updating step of the original algorithm are parallelized. The simulation experiments demonstrate that the improved PWBF algorithm provides about 0. 1 to 0. 3 dB coding gain over the original PWBF algorithm. And the improved algorithm achieves a higher convergence rate. The choice of the threshold is also discussed, which is used to determine whether a bit should be flipped during each iteration. The appropriate threshold can ensure that most error bits be flipped, and keep the right ones untouched at the same time. The improvement is particularly effective for decoding quasi-cyclic low-density paritycheck（QC-LDPC） codes.展开更多
文摘针对局部线性嵌入算法(Local Linear Embedding,LLE)短路、离群点影响大和结构信息缺乏等问题,提出基于Kernel Rank-order距离的重构权重局部线性嵌入算法(Reconstruction weight Local Linear Embedding algorithm based on Kernel Rank-order distance,KRLLE)。用核函数将样本点映射到高维使其更加线性可分,进而获得较好的近邻点集;计算重构权重系数进而得到加权重构权重,重构权重系数根据两点间相关性越大对重构贡献越大的特性来减小离群点的影响,并利用两点间的欧氏距离与测地线距离之比有效地将短路点排除在外;根据加权重构权重得到低维嵌入坐标。在ORL、Yale人脸库和MNIST手写体数据库上的实验表明,KRLLE对离群点具有更好的鲁棒性并且由于增加了结构信息,识别率得到了提高。
基金The National High Technology Research and Development Program of China (863Program) ( No2009AA01Z235,2006AA01Z263)the Research Fund of the National Mobile Communications Research Laboratory of Southeast University(No2008A10)
文摘An improved parallel weighted bit-flipping（PWBF） algorithm is presented. To accelerate the information exchanges between check nodes and variable nodes, the bit-flipping step and the check node updating step of the original algorithm are parallelized. The simulation experiments demonstrate that the improved PWBF algorithm provides about 0. 1 to 0. 3 dB coding gain over the original PWBF algorithm. And the improved algorithm achieves a higher convergence rate. The choice of the threshold is also discussed, which is used to determine whether a bit should be flipped during each iteration. The appropriate threshold can ensure that most error bits be flipped, and keep the right ones untouched at the same time. The improvement is particularly effective for decoding quasi-cyclic low-density paritycheck（QC-LDPC） codes.
