http://arxiv.org/abs/astro-ph/0212087 or the published version

HFA gives similar results to HMF, though inference of the mean seems to be more robust to initialization for the former.

Now, need to decide if there is a juicy problem that needs sophisticated dimensionality reduction with a good noise model. Is it as simple as HMF?

Give me a reference for varimax, if we are going to be using it.

I still think what we want is to SVD the lambda matrix.

can you cite HMF (Tsalmantza and Hogg) and compare it to IVW PCA (the Bailey method) at least in words. I don't need HMF to be part of this paper, but I want to see it compared.

I think this would be a good idea, sooner rather than later.

I think when we have fake data and we show the real answer, we should call it the "truth".

Can you also propagate this change through the full text?

It must be stated in the abstract, the intro, and in the method parts and in the conclusion, ie, four times, that methods like this automatically account for missing data. This obviates the need for heuristic ad-hoc patching methods like robust PCA and the like. This issue can be merged with #8

there are two meanings to this, one from vivienne wild and the like in astro, and one (totally unrelated but much cooler) from computer vision / machine learning.

Reassign this to me if you have trouble finding refs for either.