Summary: weekly mtg 20170215 (Byron, Steve, Matt, Aaron, Emily, me) - mobeets/nullSpaceControl GitHub Wiki

  • motivation: sentence on importance of using BCI - Aaron: likewise, in BCI dozens of neurons control ... - Byron: need to use BCI when mapping is unknown
  • B example: need solutions where there are multiple reasonable solutions (e.g., not 3 11 19)
  • "23 experimental sessions" update
  • maybe Unc first?
  • three bottom panels: make it bigger, less text - animals controlled two mappings: not "int" and "pert" - maybe a panel cartoon of two mappings--this is the main point - see phone photo - also add an arrow with "minutes later" under it - "all predictions made on mapping 2; some hyps use data from mapping 1 to predict during mapping 2" - maybe add panel of cartoon of how cursor would be controlled were he using the same data (maybe not on poster, but to have in back pocket)
  • something to divide hyps within column a little more, like a little line within the hist grids, or just space
  • predictions are on held out data, on the same timepoints in every hyp
  • maybe make data exactly the same for all hyps, even if it makes cld hists small?
  • emphasize that our goal is to find a way of predicting these data histograms
  • remove histogram error percentages
  • "Prediction" and "Data" text fixed
  • Results: text about number of sessions not necessary
  • Results: baseline cov error off the chart (mention in text)
  • SSS: need to pin down why cloud predicts expansion
  • Krak fig replace with cloud cartoon - steve's example with multiple mappings with same kidney shape, showing what the null dist would be in each case (n.b. no dists, just the kidneys); and then one sentence to explain this; "knowing this we could predict the null space dist for any mapping")
  • take cloud and data hists, look at range of inner 50% of hist, compare predicted range to actual range to show that we're predicting the spread well
  • put Emily ahead of Kristin