I will give an overview of the use of sampling-based approximate spectral decomposition to perform large-scale learning tasks.