Summary: weekly mtg 20170207 (Byron, Steve, me) - mobeets/nullSpaceControl GitHub Wiki

"minimal-firing" --> "zero firing" need to check best-mean hyp

  • did it really do so much better in pert. than it did in int?
    • for both monkeys?
  • if so, might want to include it in the main figs
  • can replace baseline, and move it to supp. figs

titles

  • start email chain, including emily and aaron

can move Fig 1C/1D to intro

no need to distinguish from Patrick's work with WMPs

IME: emphasize that we need to do this relative to the dimensions the subject thinks are output-null

  • maybe a sentence or two up front to foreshadow this?
  • maybe in results where we're talking about how we make predictions?
  • "note that all of the results that we are about to present hold true if we use the true mapping as opposed to our estimate"

why does turning off PCA on unc-uni still have mounds?

  • turn off bounds checking
  • then dists should have no tuning because they don't depend on row space vals
  • but there should also be NO mounds because the binning is only on potent space

Results hyps paragraphs:

  • min-energy description:

    • put muscles at motivation: "people have found in muscles that subjects prefer muscle activations that minimize..."
    • slower build into actual hyp

  • Uncontrolled: first two sentences need rephrasing

    • Steve: bring the motivation for it up front, in terms of optimal control theory
    • i.e., explicitly ground this in the literature as we are going to do with minimal-energy
    • "another finding from the motor control literature, is that activity in the null dimensions has higher variability than those that control the output."
    • overall: 1) this is what people have found in muscles, 2) this is what it would mean for neurons

  • task-transfer: first few sentences

    • overall: gentler introduction
    • not clear what "pre-existing structure" is
    • how about: "none of these previous hyps seem to be doing very well, but the one that does best is the one that use previously recorded activity. this motivates considering other hypotheses that use this."
      • hab: "first possibility is the way you structure activity in the null space depends on how you used it previously"