Summary: weekly mtg 20170125 (Byron, Steve, Matt, me) - mobeets/nullSpaceControl GitHub Wiki
our hyps are just inspired by muscle hyps
-
that's why it's hard to connect to previous work
-
so be careful about "so this is different than what was seen with muscles"
- especially since people think it's the main point/motivation of our paper
- need to word any statements like this carefully, because it is helpful
-
rename hyps
-
energy is tricky; need to say "when we say energy here, we mean..."
- if we consider only this population of cells, it could be that more spikes require more energy because they need ATP
- on the other hand, if they're part of a network, it could be that there's a baseline state, and it's deviations from that that require energy
discussion:
- maybe people should look for evidence that output-null activity is a function of the repertoire in muscles
- relate hyps to uncontrolled and habitual
titles
- "Constraints on behaviorally irrelevant neural activity"
- "Constraints on neural activity in dimensions that don't matter"
- "Neural strategies for resolving redundancy"
- "Cortical activity in the null space relies on the natural motor repertoire"
- [see photo]
orderings:
- Unc-emp after Unc-uni; "a refined version of Unc-uni"
- Then after they understand Unc-emp, we constrain it even further to get the distribution from times with the same movement
- introduce Unc-emp before talking about two mappings
- just say, to get the empirical distribution we use an earlier mapping, and then talk about it later
- can present all hyps with second mapping
- then later can say "well it turns out we had the monkey use another mapping before this, and we can use activity from this session as our empirical set"
- in supp, we have fits in reverse
- for discussion: maybe null-space selections evolve over long-term, but here we just had a monkey learn a new mapping, so maybe his strategies haven't evolved yet
- also, note that when fitting in reverse, habitual doesn't make as much mechanistic sense, but it still does well
- if we saw it break down going backwards, that might actually be more evidence for the monkey actually selecting habitually
- this could go in caption
- also, note that when fitting in reverse, habitual doesn't make as much mechanistic sense, but it still does well
Fig 1:
- A/B: should BCI output and movement be in boxes? maybe boxes aren't necessary? think about it
- B: BCI image looks weird
- maybe remove?
- B: put square around cursor/targ for monitor
- C: put hist along dotted line to correspond more to D
- C: label tick v_t
- C: remove max lines
- D: make hist look more like gauss, or something data-like
- D: put lots of smaller dots so the curve is actually true
Fig 2:
- want 3 res figs: min/bas, unu/une, hab/cld
- possibly have zoom lines to show inset is top-left col panel
- maybe just panels A and B (no D-I labels)
- bigger fonts
- maybe go back to just wedges on L column
Fig 3:
- unc-emp: write in green "estimated from previous mapping"
Errors hists fig (should be Fig 5):
- show avg across monkeys
- add all monkeys in supp
could have other fig that depending on combo of potent line and kidney bean, you might have variance expanding or not
To dos:
- title
- renaming hyps