Big news, I just accepted an offer to join Baylor College of Medicine's Dept of Neuroscience as an asst professor!
I start Fall 2024, and I'm looking for phd students, postdocs, and research scientists to join me studying learning in neural populations.
"Some people give beautiful presentations but go down in flames during the chalk talk."
🤦 Maybe that's because no one is ever allowed to see a chalk talk until they're giving their own?
Excited to share we (
@yulikeneuro
@aaronbatista
and others) have a perspectives paper on learning in neural population activity out now!
"How learning unfolds in the brain: toward an optimization view"
📎
(1/n)
Excited to share a preprint of our work "Learning is shaped by abrupt changes in neural engagement," advised by Aaron Batista, Steve Chase, and Byron Yu.
I'm (even more?) excited to finally make my own
#tweeprint
! (1/n)
Congrats to
@BenjoCowley
and team for getting this awesome paper out! They find that firing rates in V4 and PFC show a large, "slow drift" over time that correlates with animals' impulsivity.
May you never treat your trials as independent samples again.
There's lots of neuro papers studying learning, but few that I can find where neural activity was analyzed throughout learning, as opposed to just before/after.
Anyone have any favorites? (Bonus points if they don't average firing rates across neurons!)
"CS" and "US" in classical conditioning has to be the worst pair of acronyms ever agreed upon. It takes conscious attention for me to figure out which is which every...single...time
Which line do you think does a better job of predicting Y given X?
(Hint: This is one of those days where very basic statistics confuses the hell out of me.)
(1/n)
"Learning leaves a memory trace in motor cortex"
Out today in Current Biology! Led by Darby Losey, with myself, Emily Oby,
@MattGolub_Neuro
,
@aaronbatista
,
@YuLikeNeuro
, Steve Chase, and team.
📎:
🧵:
"Learning alters neural activity to simultaneously support memory and action"
Excited to share a new preprint, led by Darby Losey, with myself, Emily Oby, Aaron Batista, Byron Yu, Steve Chase, and team. 🧵
"Learning alters neural activity to simultaneously support memory and action"
Excited to share a new preprint, led by Darby Losey, with myself, Emily Oby, Aaron Batista, Byron Yu, Steve Chase, and team. 🧵
Take a population of neurons, and then sort them based on when their peak firing rates happen in time.
- are there agreed on statistics for quantifying this temporal structure?
- anyone know the ref for that paper about the importance of cross-validation when making these plots?
Our paper on probabilistic representations in RNNs during reinforcement learning is now up at
@PLOSCompBiol
! The reviewers made some really great suggestions that I think improved the paper a lot since the preprint.
It's time for the 1st annual
#tweeprint
census! 🤓📊
Since June of this year, I've been collecting as many neuroscience and neuro-related tweeprints on
@tweeprint
as I could find. I just analyzed all 251 of them, and wanted to share what I found. (1/n)
Some fun life news I haven't shared yet: I've moved to the Boston area and will be starting a postdoc in August with
@gershbrain
!
Really excited to learn about dopamine/RL, but also to have a good reason to tell my friends/family why dopamine squirts aren't about pleasure
Since today is representational drift day, I just wanted to share my two cents, as this is something I've been thinking a lot about in terms of population activity. (1/n)
If you're registered for
#cosyne2021
tomorrow, come find out how internal states influence learning in population activity!
I'll be there for sessions 1a and 1b (
#1
-077)
My kids, last month:
- learned to crawl and to stand
- tasted many foods for the first time
- learned to recognize words
Me, last month:
- sat in one place for hours at a time, every day
- ate oatmeal for breakfast every morning b/c it's easier
- forgot my wifi password
It's finally out! I'm thrilled and excited to share this tour-de-force work on [topic], led by the talented [first author]. This was truly a team effort, including the twitterless [oldest PI].
Check out the paper here [link], or see thread 🧵
Today I turn 33. The #'s 33, 34, and 35 are each semi-prime because they are the product of exactly 2 primes. Because it's the longest run of semi-primes I am likely to live through, as of today I am officially in my semi-prime.
(Sorry, I'm a dad, I can make these jokes now.)
@grassmannian
If you want to evenly distribute N points on the surface of a sphere, you can only do this exactly if N = 4,6,8,12, or 20. I don't know why but this fact just disturbs the hell out of me.
Really excited that my first first-author paper is now out
@eLife
! We looked at how the brain selects which neural activity patterns to generate when there are multiple (redundant) options for effective control.
Matlab users: If you're tired of manually installing things from File Exchange (download, unzip, add to path, forget it exists, repeat), try out Matlab Package Manager!
Installing something can be as easy as:
>> mpm install export_fig
Whenever I'm struggling with writing, it feels like I know what I want to write but the words just aren't there. But then, once I've figured it out, I realize it was immature/foggy ideas that were the problem, not the words. Anyone else have this experience?
Traveling before kids: buy a plane ticket, sleep on a friend's couch
Traveling after kids: buy multiple plane tickets, find an airbnb, rent a car, rent cribs, rent car seats, buy groceries, entertain tired children in a place that's missing most of their favorite toys
Here's one example from
@DurstewitzLab
, looking at abrupt changes in population activity in PFC while rats learned a new rule: "Abrupt Transitions between Prefrontal Neural Ensemble States Accompany Behavioral Transitions during Rule Learning"
We've posted a new paper on dimensionality reduction of calcium imaging recordings. This work was led by Tze Hui Koh, jointly with Steve Chase, Misha Ahrens, Will Bishop, and team.
This is so cool! I'd always figured (naively) that a neuron's dendrite branching pattern was relatively random, but clearly not the case if there are neurons with basically a mirror opposite structure!
We found that most motor neurons had unique and recognizable branching patterns. This let us find left-right pairs of motor neurons that likely control the same muscle in opposite legs. Identifiability of motor neurons is a key advantage to studying motor control in the fly. 6/n
TIL: Let's say you've got 100-d data (e.g., firing rates of 100 neurons), and you want to find a dimension along which the activity is maximally autocorrelated.
Surprisingly, even if your data is pure noise (i.e., iid Gaussian), you can expect to find a dimension with ρ ≈ 0.4.
The bare truth of science:
1. Nobody believes a computational model except the person who built it.
2. Everybody believes an experimental measurement except the person who made it.
Found this book in a conference room and was very excited by how creepy it looked. But it turned out to be super interesting! Written by a Stanford sleep researcher in the 70s. Turns out the author a) coined the term "REM," b) identified/named the five stages of sleep, ...
Starting to look for a postdoc position, and I'm realizing how little I understand about how the academic world works. Everything seems so strategic and yet totally up to chance.
Anyone have any good tips/resources for how to pick where to do a postdoc?
This is my daycare costs for 2 children in Seattle during the month of July. Please tell me how grad students and postdocs are able to afford childcare.
As opposed to what??
Popular books and articles about the brain often promulgate a form of Cartesian dualism, making it sound surprising that mental events are produced by brain events.
Very excited to report that my toddlers are clearly Bayesians 🧐
When I ask them "what color is this?", they always guess "red" first, then "blue," then "green", apparently because those are the most likely answers based on my past questions.
It's so ridiculous how awful/non-existent retirement options are as a grad student and postdoc. Our pay is low enough, can't we at least get some decent retirement options?
So grateful that in my 2.5 years as a postdoctoral fellow, my employer invested in my retirement plan.
Receiving this gives me a real sense of financial security for the future!
🤔
Finally, if you're interested, the (slightly incomplete) list of tweeprints I've collected is here:
Also, please let me know if I have missed your tweeprint! (n/n)
Anyway, just wanted to share my thoughts--I think neural drift is a super interesting feature of population activity, and one that has the potential to answer lots of really interesting questions about the brain. (n/n)
3yo woke up sick, refused to eat/drink anything all day until about 4pm. At that point we thought, ER here we come. I tried one last thing: "Did you know every time you drink water it makes mommy make a funny noise?" He then drank all his water, and is apparently totally fine
Just had our kids tested for lead in their blood, and one initially tested above the 5 mcg/dl threshold.
Pittsburgh is a hotspot for this sort of thing. Below, the % of kids who tested above 5 mcg/dl. National average is "less than 5%".
Thread on why this is fucked (1/n)
@bcaessens
@stevenstrogatz
Let a=“xy”, where “x” and “y” are the first and second digit of a. And b=“wz”. Now add these like you learned in elementary school, where you add the ones column and then the tens
In the ones col we know y+z=10, so we carry the 1. This means w+x+1=10 in the tens. Thus x+y+w+z=19
The top 3 most-liked tweeprints from 2019:
1. Weight Agnostic NNs
@hardmaru
2. High precision coding in mouse V1
@computingnature
3. Cortical Neuron as a Deep Artificial Neural Net
@DavidBeniaguev
(7/n)
A question the field should have been asking from the get-go: If you train an RNN to fit empirical neural data/behavior, can you really interpret the RNN's dynamics as representative of the empirical dynamics?
Excellent work from
@CPehlevan
's lab!
Super excited to see this work published! Congrats to Emily for finishing such an awesome project. These were super difficult experiments for her to run, and a really challenging problem to think through!
@MattGolub_Neuro
@AlanDegenhart
Repeated addition is called multiplication, repeated multiplication is called exponentiation, and repeated exponentiation is called...tetration!
And visualizing it gets you this pretty picture. (Found via
@johncarlosbaez
, my new favorite twitter account)
At first I was skeptical of the idea of studying dexterity in rodents. But this video! My four-month-olds still can't do what this mouse is doing!
(From a great talk today by
@AdamHantman
on the neural circuits of dexterity--including some awesome work by
@bsauerbrei1
)
One thing that surprised me about having kids is how time gets recalibrated. A year used to feel like no time at all to me, but now I think about how much my kids will change in a year and it blows my mind!
1) The inflexibility of neural variability throughout learning.
2) The brain's use of multiple learning processes, even for simple tasks.
3) Changes in neural activity that occur without regard to the task at hand, like changes due to synaptic turnover or internal state. (5/n)
1. Work by
@XuLunaSun
finds gradual drift in motor cortical population activity during learning, and they see drift even for targets that *didn't* need to be learned. In this case, neural drift may be a way of "indexing" different motor memories (5/n)
Also a huge thanks to
@lucylai_
,
@JeffMYau
, and
@NeuroPolarbear
for making this job happen and giving me so much advice on my application and job/chalk talks! I can't wait to talk science!
2. Our work in BCI learning also finds drift in motor cortical pop. activity during learning, and here it persists even when it *hurts* performance. In this case, drift is *not* learning--it looks more like the animal's arousal/engagement.
(6/n)
The end of Amy's Science paper made such an intriguing proposal: Maybe animal's choices are aligned w/ noise variability dimensions b/c they're using a decoder that's "general" rather than specific to the cues used in the task. So excited to see this fleshed out in a full paper!
New paper! A very fun collaboration between
@amymni
and
@cc_huang11
(with
@BrentDoiron
and me along for the ride) was just published in
@eLife
“A general decoding strategy explains the relationship between behavior and correlated variability” 🧵1/9
Before our babies were born, Jess and I spent two yrs making music together. We put in a lot of time on these, so I’m really excited to finally share them! You can listen in full at . Let us know what you think! The video is some creepy/sloppy StyleGAN.
Overall, we suggest that to explain changes in neural activity during learning in terms of optimization, we should consider how ANNs and brains differ. We hope this will lead to better models of how neural activity changes during learning, and possibly better AI as well. (n/n)
@neuroecology
The funniest part of this is how at the end of explaining general intelligence and genetics and everything else, the last sentence says something like “the problem that remains is understanding photosynthesis”.
Me, at Fedex: Hi, I'd like to mail this document.
Fedex guy: Sir...We do not "mail", we SHIP.
Me: Okay, I'd like to...ship this document?
Him: Sure, I can help you with that.
Where are most tweeprinted papers published? 136 of 251 tweeprinted papers were on a preprint server (93 on biorxiv). What really surprised me here though was how many tweeprinted papers were published in higher-impact journals such as nature/cell/elife... (3/n)
Great write-up on my recent
@eLife
paper on neural redundancy. Their interpretation is that our results are consistent with M1 being controlled (e.g., by Pmd) subject to a minimal energy constraint. Very interesting to think about M1 as the plant and not as the controller!
Ta-Chu Kao and Guillaume Hennequin discuss recent work by Steven Chase and colleagues in a new Spotlight article "Null Ain’t Dull: New Perspectives on Motor Cortex"
Probably one of the best reasons more scientists should get on twitter is the exposure to all the bullshit that women in science have to deal with. I’ve learned so much from reading threads like this.
Friends at conferences - please do not assume that the people that you talk to do not know anything. I just got told that I should read what Stanton et al found about pain.
I. Am. Stanton.
Here's how neural activity varied trial-to-trial for each of 8 targets (each dot is a trial, color indicates target), before monkeys started learning a new BCI mapping. The orange lines summarize this variability. We're calling them "neural engagement axes."
(6/n)