Summary: weekly mtg 20161207 (Byron, Steve, Matt, me) - mobeets/nullSpaceControl GitHub Wiki
Discussed:
histogram overlap
- just values at first
- values for particular session in figure
- and then in text "on any session, no hyp did better than __"
- then in supp, show bars for first half
- also in supp show scores for each session (heatmap plot)
- scores for each session for first part alone too!
make it clear there are multiple datasets
- for first half, use intuitive
- for second half, make it super clear that we can now refit the previous hyps now on different data, to make it easy to compare with the new hyps
SSS: try cloud with intuitive row space values to make expansion-contraction values
- does this make cloud have ratio nearer 0? if so, this explains why normal cloud has value > 0: just due to resampling given different freqs of row space vals
SSS by monkey?
supp figs: fits in reverse
check how often cloud samples different points
- i.e., does cloud force you to visit every point in the space?
take intuitive data and view in perturbation histograms
- this will tell us if distributions are really preserved
- and then emphasize that cloud does predict a slight distribution change, and in the right direction
- because otherwise, why do we need the simulations?
- i.e.: view intuitive, observed (perturbation), and cloud predictions
add discussion paragraph on whether distributions change
- and whether or not this is a version of unconstrained
title: behaviorally equivalent
- keep this in abstract
- only mention output-null once we get into text, and make an explicit link to behaviorally equivalent
- i.e., "the behaviorally equivalent points all lie in the same output-null space"
NDSEG:
- change phrasing to make stronger claims
- e.g., "I am going to BE THE GUY who SOLVES THIS PROBLEM"
- also, make sure there are two sentences of "deep dives" in third para
- as opposed to just high-level
To discuss:
- title: Predicting neural activity in [behaviorally OR task] [irrelevant OR equivalent] dimensions
- text: task-equivalent vs. output-null
- text: "muscle-inspired" vs. "energy hypotheses"; "hypotheses in multiple tasks" vs. "task-transfer hypotheses"
- marginal distribution error metric
- supplemental figures
- Fig 4: put behavior in supplement; describe IME after showing results?
- SSS: error bars; why cloud a slight expansion?
Tiny things:
- author contributions
- use Nelson data --> Emily as author?
- highlights (bullet points)
- in brief (one sentence of < 40 words)