|
Abstract:
|
Functional embedding aims at dimension reduction for large p, small n data by ordering the elements of the p-dimensional predictor vectors via univariate scaling, applied to correlation-based dissimilarities. Under regularity conditions, the embedding maps predictor vectors to a smooth random function for each subject, subsequently represented by just a few functional principal component scores. For high-dimensional gene expression data as predictors, functional embedding coupled with functional regression procedures outperforms previous methods for the classification of leukemia and for the prediction of survival time of lymphoma patients. Talk based on collaborations with Ping-Shi Wu, as well as with Kun Chen and Jane-Ling Wang.
|
