我院2020级硕士研究生张晓文同学以第一作者，指导教师陈争光教授为通讯作者在中科院二区期刊《SpectrochimicaActa Part A: Molecular and BiomolecularSpectroscopy》在线发表了题为“Applicationof Adaptive Laplacian Eigenmaps in Near Infrared SpectralModeling”的研究论文（DOI:https://doi.org/10.1016/j.saa.2022.121630，文章链接：Applicationof Adaptive Laplacian Eigenmaps in Near Infrared Spectral Modeling -ScienceDirect）。
LaplacianEigenmaps is a nonlinear dimensionality reduction algorithm based ongraph theory. The algorithm adopted the Gaussian function to measurethe affinity between a pair of points in the adjacency graph.However, the scaling parameter σ in the Gaussian function is ahyper-parameter tuned empirically. Once the value of σ is determinedand fixed, the weight between two points depends wholly on theEuclidian distance between them, which is not suitable formulti-scale sample sets. To optimize the weight between two points inthe adjacency graph and make the weight reflect the scale informationof different sample sets, an adaptive LE improved algorithm is usedin this paper. Considering the influence of adjacent sample pointsand multi-scale data, the Euclidean distance between the k-th nearestsample point to sample point xi is used as the local scalingparameter σi of xi, instead of using a single scaling parameter σ.The efficiency of the algorithm is testified by applying on twopublic near-infrared data sets. LE-SVR and ALE-SVR models areestablished after LE and ALE dimension reduction of SNV preprocesseddata sets. Compared with the LE-SVR model, the R2 and RPD of theALE-SVR model established on the two data sets are improved, whileRMSE is decreased, indicating that the prediction effect andstability of the regression model are established by the ALEalgorithm are better than that of the traditional LE algorithm.Experiments show that the ALE algorithm can achieve a betterdimensionality reduction effect than the LE algorithm.