Title: Stringing via manifold learning: Improving the functional representation of high-dimensional data
Authors: Harold A Hernandez - Universidad Carlos III de Madrid (Spain) [presenting]
Rosa Lillo - Universidad Carlos III de Madrid (Spain)
M Carmen Aguilera-Morillo - Universidad Carlos III de Madrid (Spain)
Abstract: Stringing via manifold learning is discussed. This method maps general high-dimensional data to functional data. It assumes that the sample vectors are realizations of a smooth stochastic process, observed with a scrambled order of its components. The key ingredient, a reordering step, is fundamental before recovering the true underlying process generating the data. The earlier version of stringing (based on multidimensional scaling) is improved by incorporating manifold learning. With this proposal it is possible to recover non-linear relationships between predictors, resulting in a better reordering of the data and a more reliable functional representation. Through simulation studies it is shown that the proposed method outperforms the original approach. The potential of using stringing and functional modeling in the high-dimensional scenario is also addressed. Real data applications on gene expression and single-nucleotide polymorphisms (SNPs) arrays are presented.