Feedback: Doug (20170222) - mobeets/nullSpaceControl GitHub Wiki


Doug feedback

  1. Intro

    • Intro feels like it misdirects by talking about muscle literature
    • feels a little long-winded
    • could cite the muscles when you introduce the hyps, but maybe not in the intro
    • just start up front with what you're interested in: neural redundancy; it's been unaddressed so far, understanding the selection process is important, etc.
    • seems like a paragraph too long
    • "We initially considered" <-- Yes. Should be second paragraph.
    • "unknown whether the results regarding muscle literature" <-- what results?
    • "different mapping" totally unclear at this point; or just "mapping"
      • also can reference 1B here
    • "each type of cursor movement" <-- unclear
    • "the brain can produce only a limited..."
      • maybe focus on M1? touch on this in discussion
  2. Velocity vs. direction

    • Fig. 1: hung up on monkey really wanting v=4 vs. v=2, for example
      • direction makes much more sense
      • maybe just remove velocity? to be more vague
      • also, text refers to cursor-vel axis as "output-potent"
      • maybe no reason to have two histogram traces in C,D
    • don't hear about "cursor movement direction" until page 9
      • should talk about this earlier, say in intro
  • "choose"/"select"/"decision"/etc.

    • pick a word and stick with it
  • Doug: when he reads about FA or preprocessing, he wants to know what it would be like without it

    • need a sentence that explains why we needed to do this
    • otherwise, doesn't understand the impacts this will have on the results
    • also be more explicit about the fact that this isn't just a preprocessing, this is actually how the experiments were run; this will also help the reader believe you
  • discussion point:

    • trial dynamics might explain a lot; difficult to go from UP to DOWN, e.g.
  • Results:

    • say number of monkeys
    • center-out task: no explanation, no citation, no reference to methods
    • typo: "we made be able to"
    • "previous" vs. "different" --> important thing is that it's "different" right?
    • "estimating the error...does not capture..." <- didn't understand what this was about (motivation for using mean/cov error)
    • IME part: seemed undermotivated, and possibly undersold
  • Fig. 2:

    • explain why you're showing only the top 3 dims
    • would be nice to give better intuition for how the prediction is made
    • feels like there could be some sort of better way to explain quantitatively why these fits aren't good; you just have to wait til later figs...
      • maybe add a table?
  • Fig. 6

    • need units on A
    • no ref to B in the text
    • C: need help interpreting this figure
      • why does MnkN look messy?
      • need more hand-holding
  • Discussion:

    • first para: yeah it's different than muscles, but what do we learn about M1?
    • how population size, spatial clustering of neurons affects range of activity that's possible in these spaces (e.g. this is just from a Utah array)
    • why was BCI the right way to do this experiment? not totally sold that you had to, so really nail it in
    • the "how" and "why" the brain is "choosing" in this way is unclear
    • what's the cartoon model for how this space looks like, when the monkey needs to move up, say? is the animal actively going to a remembered state that's likely to work?
  • Methods: seems really good

  • Liked title: Predicting the selection of behaviorally

    • likes it better than redundancy