摘要
Hessian LLE算法是一种经典的流形学习算法,但该方法是以批处理的方式进行的,当新的数据点加入时,必须重新运行整个算法,计算所有数据点低维嵌入,原来的运算结果被全部丢弃。鉴于此,提出了一种保持局部邻域关系的增量Hessian LLE(LIHLLE)算法,该方法通过保证流形新增样本点在原空间和嵌入空间局部邻域的线性关系不变,用其已有邻域点的低维坐标线性表示新增样本点,来得到新增点的低维嵌入,实现增量学习。在Swiss roll withhole和frey_rawface数据集上的实验表明,该方法简便、有效可行。
Hessian LLE algorithm is a classical manifold learning algorithm.However,Hessian LLE is a batch mode.If only new samples are observed,the whole algorithm must run repeatedly and all the former computational results are discarded.So,incremental Hessian LLE(LIHLLE) algorithm was proposed,which preserves local neighborhood relationship between the original space and the embedding space.New sample points were linearly reconstructed with exis-ting embedding results of local neighborhood samples.The proposed method can learn manifold in an incremental way.Simulation results in Swiss roll with hole and frey_rawface database testify the efficiency and accuracy of the proposed algorithms.
出处
《计算机科学》
CSCD
北大核心
2012年第4期217-219,226,共4页
Computer Science
基金
国家自然科学基金资助项目(60875040
60970014
61175067)
教育部高等学校博士点基金(200801080006)
山西省自然科学基金资助项目(2010011021-1)
山西省科技攻关项目(20110321027-02)
太原市科技局明星专项(09121001)资助