Finding high-order correlations in high-dimensional biological data

Abstract

In many emerging real-life problems, the number of dimensions in the data sets can be from thousands to millions. The large number of features poses great challenge to existing high-dimensional data analysis methods. One particular issue is that the latent patterns may only exist in subspaces of the full-dimensional space. In this chapter, we discuss the problem of finding correlations hidden in feature subspaces. Both linear and nonlinear cases will be discussed. We present efficient algorithms for finding such correlated feature subsets.