"Multilayer" spectral embedding reference

[maxentile]

I've only skimmed this so far Learning Eigenfunctions Links Spectral Embedding and Kernel PCA (Bengio et al., 2004) but it looks really useful.

They describe connections between many manifold learning algorithms and explain out-of-sample extension of the learned embeddings. At the end they also suggest to look into multiple iterations of such feature transformations, although I'm still having trouble finding a reference that actually does it.

[maxentile]

Oh wow this is also really interesting: Kernel methods for deep learning