Summary: weekly mtg 20160810 (Byron, Steve, Aaron, Matt, me) - mobeets/nullSpaceControl GitHub Wiki
Discussed
Fig. 1
B.
- add a funnel? funnel on the BCI mapping? in the BCI mapping? make the box itself a funnel?
- box itself a funnel, wide arrow coming in, tiny arrow coming out
- make box flow vertical (also add in array, above monkey's head)
- make a straighter dotted trajectory
- lower monkey so he's looking at the screen
- add more units
Fig. 2
A.
- have dotted max lines intersect the axes, potentially label them as max; maybe they're not a perfect box
- get rid of the gray cloud
- get rid of black points?
B-D.
- make sure to mention in text that we're binning output behavior
- maybe try with rectangles/squares?
- make them thinner, and say "consider three bins" rather than them spanning all outputs
Fig. 3
A.
- draw the screen
- to distinguish cursor vs. center
- also, target doesn't lie on outer ring
- make sure the dotted line lines up with the vector
- can this fit in Fig. 1?
B-D.
- skip
Fig. 4
A.
- make x-labels into titles
- show where 0 is, give some ticks
B.
- pick a different dim? need to make tuning clear
- make x-labels into titles
- show where 0 is, give some ticks
- maybe pack stuff into the x-lim, e.g., find 95% max
- 8x2 instead of 4x4?
C. stand-alone figure
Fig. 5
B.
- arrows not clear from rotation
- make ellipses longer so it's clear they're not rotations?
- maybe find new way of viewing hypotheses?
C. this one is more effective
Fig. 8
A.
- order from best to worst
- pick more distinct colors
- can also rescale vertical axis so we can see all
QUESTION:
- source of baseline mean?
- orthonormalized vs. spikes/bin vs. ordered columns
- orthonormalized and SCALED
- SVD/sort on null space
- okay, so just do PCA on observed null space activity for displaying/evaluating
TO DO:
- use a TRUE Uncontrolled a la baseline/minimum that has a uniform distribution
- do a similar improvement on minimum/baseline, where it uses true data
- check hypotheses in SSS, for detcov vs. learning
- supplemental fig for performance under two mappings, and with IME (similar to last fig of matt's elife paper)
- supplemental fig for where behavior converged