Summary: weekly mtg 20160713 (Byron, Steve, Matt, Aaron, me) - mobeets/nullSpaceControl GitHub Wiki

try the stillIrrelevant vs. nowIrrelevant with the same number of dims

• try doing this in reverse too
• and check out histograms

figure for nowIrrelevant result

dimensions that are in row space of intuitive have selection bias: maybe this has something to do with these being more volitionally controllable, that makes this nowIrrelevant error very low

minimal tuning curve change hypothesis: single neuron preserved tuning curve hypothesis?

• (tuning curve as fcn of target location)
• is this just habitual with spikes?
• but of course the tuning curves change; they just give us an impoverished view of...blah blah. consider mentioning why we don't consider this hypothesis.

how to normalize covariance?

• by total shared variance

Fig 3B not helpful

• Maybe make 4Ai and marginal histograms as 3B

Fig 3: minimal, unconstrained, baseline

• a. marginal histograms
• b. error in mean
• c. error in covariance
• d. error as a function of theta
  • check the histograms, does this really give us anything the histograms don't? (except for unconstrained)

In Introduction disclose everything. Fig 3: minimum/baseline/unconstrained Fig 4: habitual/cloud

So for those dimensions that switch from being relevant to being irrelevant, what happens?

Maybe retune figures 2 (cartoons) to focus on this activity that was relevant but is now not

• just plot overlap of intuitive and perturbation activity in that stillIrrelevant space

still up:

• present in terms of spikes or factors?
  • send email about this