Summary: weekly mtg 20170222 (Byron, Steve, Matt, me) - mobeets/nullSpaceControl GitHub Wiki

Acknowledgements: add Wilsaan, Doug, anyone else who's significantly contributed to how we think about this

in text (and caption), emphasize that we're showing velocity, but really we looked at similar cursor directions
- maybe also move to output-potent
"selected"
in discussion, bring up that we ignored dynamics, but still did pretty good
for hyp hist errors in text, maybe acknowledge that i'm leaving it open, and will compare it to others later
- just be more agnostic when reporting the quantitative value
IME: maybe this should come when we talk about what is behaviorally equivalent to the animal
- monkey has some interal idea of what the mapping is, which might not match with what it actually is, so we were careful to make sure we were estimating what that method was. we recently developed a method to do this, and we apply that here. we also did it with the real mapping and the result don't change
discussion:
- need to talk timescales
  - need to cite the strick paper for minimal firing not necessarily meaning minimal energy for metabolics
- activity might be handled differently based on a neuron's output, e.g., we don't know which neurons are corticospinal tract, for example
cloud: distribution vs. sampling procedure
- how similar are these?
- what's the hist error of one vs. the other
cloud: sample from uniform distribution
- how: take 200 nearest pts in potent space, find min and max across all dimensions, sample uniformly
- why? because there could be unequal sampling of potent space between training and test (e.g., in intuitive he used one cursor velocity more than in perturbation)
unc-uni: need to mention in text why they're not boxy
- add that supp fig (photo in phone)
best-mean firing: present instead of baseline
- could mention in methods, say, that it's really close to the baseline rate
- but otherwise, don't need to mention the baseline hypothesis