Notes from second year milestone presentation (20170821) - mobeets/nullSpaceControl GitHub Wiki

Rob Kass's comments:

effect of choice of dimensionality; e.g., what if you had 12 factor dimensions?

how stable are the potent dimensions? i.e., if you had 12 dims, would you choose the same 2?

for fixed repertoire, convey conception and not implementation; e.g., specify distribution and not just support

is the fixed repertoire as good as we can do, in terms of noise? e.g., sometimes with noisy data a linear model and nonlinear model are not distinguishable due to the variability. similarly, maybe other hyps can't be better than fixed rep. with this approach.