ABSTRACT
In some applications, it is insufficient to model the data using linear models directly. In this chapter we present nonlinear dimensionality reduction algorithms for multilabel learning using the kernel methods [207]. The basic idea of kernel methods is to map the data instances from the original space to a high-dimensional Hilbert space (feature space) such that linear models in the feature space correspond to nonlinear models in the original input space. We use the Drosophila gene expression pattern image annotation problem as an application example for the kernel-based dimensionality reduction methods.
