Scabby feedback on 20170120 - mobeets/nullSpaceControl GitHub Wiki

High-level comments

  • motivation: how does this help us know about movements
    • ben: focused more on the reassociation
  • will: min-energy reasonable interpretation in neurons?
    • refs for interpreting that in neurons?
      • maybe fMRI?
    • need to help guard against straw man
    • steve: can pry find more papers that make this assumption
    • Ryan: "minimal energy" isn't necessarily suppressing activity at a network level, as it might take more energy to inhibit activity, e.g.
      • also, cloud could be seen as min-energy, as maybe habitual activity points are easier to generate
  • sharlene: results still felt like intro
    • maybe condense down, or choose one
  • emily: hung up on claim that a diff. mapping is a diff. task
    • motivations and goals are the same
    • mostly a semantics thing, with the word "task"
  • alan: struggle with reliance of cloud on having a 2nd task
    • seems like a different problem in a way; not just trying to predict activity like the original hyps do
    • make the link stronger, why we should care
    • emily: could we compare activity between BCI and arm control? i.e., could you use BCI activity to predict arm control activity, or vice versa?
    • steve: might mention is discussion how it's not known how much the repertoire might change if the task/context varies a lot,

Confusions

  • "predicting" neural activity; make it clear what we're predicting vs. how we're predicting it
    • Ben: "Are you predicting neural activity?"
  • scott: didn't understand what the distributions were
    • emily: still struggled to interpret histograms
      • need units (spikes per timebin)
      • maybe label null dim 1, e.g.
  • ryan: trouble linking the first "three" hyps mentioned in abstract/intro, but didn't see a particular fig linking those three together
    • e.g., unclear how to tie hyps in intro to hyps in figs

Figs

Fig. 1:

  • array is upside down [Scott]
  • arm not connected to the head [Emily]
  • mouth creepy Joker smile [Emily]
  • not enough parallels with arrows in human vs. monkey [Emily, Ben]
    • could have numbers for scale (e.g., # of muscles, neurons, etc.)
    • replace black arrows
  • feels a little empty; could work hypotheses into this
  • might emphasize the redundancy part (the funnels) somehow rather than the drawings
  • Aaron summary: want to understand yellow arrow in A; to do that we need to understand yellow arrow in B; hyps motivated by muscles->movements studies in A.

Fig. 2:

  • could probably be part of Fig. 1, as this is still
    • might think again about a 3D version where we label IM, control axis, null axis; then zoom in say now we're just focusing on this 2D space
    • so make 3D version Fig. 1C, and 2D version Fig. 1D
    • the tricky part is neurons/factors differences
  • two dots not three
  • make it clearer that the dots have the same output-potent value
    • maybe have the dots have the same color, but the axis to have a different color
  • how to get rid of more lines?
    • remove max neural firing lines

Fig. 3:

  • TOP: overarching titles of hyps, e.g., "minimal energy"
  • BOTTOM: too overwhelming how many panels there are
    • maybe focus on one panel of histograms, and make this very clear
    • then maybe present the others as in Fig. 7B
  • data dist (black curve) should be labelled
  • show more distinct directions rather than adjacent ones
  • no purple in uncontrolled
  • label "directions" over the circle panels
    • also, make sure it's divided into 8 and not 16
  • maybe work a quantification in there somewhere
    • if you focus on one panel

Fig. 4:

  • remove "A" and "B"
  • make it clear that B rotates the axes of A
  • "mapping #1" on axes isn't that helpful
    • maybe just use color here to distinguish
  • text of "repertoire" is helpful

Fig. 5:

  • trouble distinguishing between Hab and Unc [Will, Emily, others]
  • would be cool to highlight that two of the hyps are making predictions that weren't observed before [Scott]
  • "is cloud predicting that monkey is restricted?"
    • maybe just choose a different strip
  • unc-empirical is confusing
    • might want to put it next to unc-unif, not hab, to make it easier to understand
  • maybe just swap B and A so unc comes first
  • confusion: text made it seem like all the green dots should have same output-potent value in mapping one [Scott/Rudina]
  • might one to find a way of depicting unc-emp that has nothing to do with "corrections" under learning [Steve, Scott]

Fig. 6:

  • use broken axis for 6E to show baseline
  • graphically, just depict difference between cloud and anything that looks close
    • but still say in text that they're all sig. different
    • but nevermind, it'll get too confusing, it's just messy

Fig. 7:

  • more specific in B
  • could consider highlighting the shading between the ellipses
  • mini-panel axis labels?
  • in main text, emphasize the importance more: "These are the dimensions you might think would definitely change."
  • make sure to rotate so cursor-directions are along y-axis not x-axis

TEXT:

  • could have transitioned better
    • from the prev hyps to the next ones
    • seems like "now for something completely different"
  • "overlap" undersells because it makes it sound like we're just looking at the support [Will]
    • Emily: felt like shorthand; give it a name that's more intuitive
    • just say "different in distributions, per dimension, per direction"
  • referring to the "new" hypotheses vs. "old" hypotheses not helpful--make the names of the groupings clearer or at least refer to them more consistently
    • if the task-corrections one share some critical aspect, emphasize that more