摘要
针对局部线性嵌入(LLE)降维算法中邻域参数的人工设定及其全局性的缺陷,研究了聚类和降维的内在联系,提出了邻域参数的自适应选取策略,构建了一种聚类和降维的自适应局部线性嵌入(ALLE)算法,为每个样本点设计最佳的近邻搜索空间,自适应选取邻近点计算权值重建矩阵,基于类信息重新定义了重构误差函数。实验表明,新算法更能体现出数学上流形概念的局部坐标化本质,对不相关数据、冗余数据和噪声数据具有良好的鲁棒性,在实际识别问题中体现出优越的性能。
In order to improve the traditional locally linear embedding（LLE）,a strategy to select parameters adaptively is proposed by studying the relationship between clustering and dimension reduction.An unified computation model for simultaneous clustering and dimension reduction is proposed.The novel adaptive algorithm designs the optimal neighbor space for each sample.It’s adaptive to select adjacent points for calculating reconstruction matrix.And the error function has been redefined based on categorical information.Compared with LLE algorithm,the presented algorithm preserves the local-coordinates of the manifold more efficient,and it is robust to irrelevant,redundant and noise data.Experiments demonstrate that this algorithm has a good performance on classification.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2010年第5期772-778,共7页
Journal of Optoelectronics·Laser
基金
国家自然科学基金资助项目(60975015)
重庆市科技攻关资助项目(CSTC2009AC2057)
关键词
局部线性嵌入
自适应参数
最近邻传播聚类
重构误差
locally linear embedding（LLE）
adaptive parameter
affinity propagation clustering
reconstruction error
