Lightnews — Scholar-powered news

deanpospisil.bsky.social @deanpospisil.bsky.social · Aug 19

Thrilled to join UIUC as Prof of
Computational Neuroscience! Lets figure the brain out before we're dead! Recruiting PhDs via CS, ECE & Psych. We study stats methods, mechanistic model inference (w/ connectomics), sensory coding & more. Reach out! [email protected]

2 8

Reposted

Dr Craig R McClain @drcraigmc.bsky.social · Jun 24

I've wanted to write this article for years. About my and other's struggles to even survive sometime in #academia. Thank you to the amazing editors at @plosbiology.org that gave me the forum to write this piece. #science

Too poor to science: How wealth determines who succeeds in STEM

From student to researcher, a career in science can come with a high price tag. This Perspective explores how persistent financial barriers limit who can succeed in science, revealing how wealth shape...

journals.plos.org

20 180 350

Reposted

Rich Pang @rkp-science.bsky.social · May 27

How do we get more neuroscience out of our behavioral data? Excited to share new work with C.A.Baker, M.Murthy and @jpillowtime.bsky.social, where we use natural behavior data to extend predictions from neural recordings about population codes for dynamic social stimuli: tinyurl.com/2d3wwfyf

Inferring neural population codes for Drosophila acoustic communication | PNAS

Social communication between animals is often mediated by sequences of acoustic signals, sometimes spanning long timescales. How auditory neural ci...

tinyurl.com

3 10 26

deanpospisil.bsky.social @deanpospisil.bsky.social · Jun 16

Someone please do this with a neuroscience foundation model. Start with known computations: reichardt detector in fly, inter-aural delay lines in owl, etc. @tyrellturing.bsky.social ?

Ted Underwood @tedunderwood.com · Jun 12

A different approach to interpretability. Instead of probing neurons and circuits, *tune* the models to provide "accurate, quantitative descriptions of their own internal processes during certain kinds of decision-making," and then show that this ability generalizes. #MLSky 🤖

Self-Interpretability: LLMs Can Describe Complex Internal Processes that Drive Their Decisions, and Improve with Training

We have only limited understanding of how and why large language models (LLMs) respond in the ways that they do. Their neural networks have proven challenging to interpret, and we are only beginning t...

arxiv.org

2 2

deanpospisil.bsky.social @deanpospisil.bsky.social · Jun 6

No feedback from cortex. Within the retina yes.
Would you agree the whole point of opsins is to rep. photon count weighted by wavelength?
I want to understand at what point you would say something is a representation. Not sure how to convince.

1

deanpospisil.bsky.social @deanpospisil.bsky.social · Jun 6

Or do you prefer the word encode?

deanpospisil.bsky.social @deanpospisil.bsky.social · Jun 6

Photoreceptors definitely represent photons. To hold off on that judgement until we have a complete theory of the brain seems impractical. Similar arguments could be made for other sensory systems, especially early. Are you thinking mainly of higher order areas?

2

Reposted

Ben de Bivort @debivort.bsky.social · May 28

Believe it a not, a third multisite connectomic project also lost funding. The multi-PI R01 led by @darbly.bsky.social with myself and @bassemh.bsky.social as co-PIs.

The Transmitter @thetransmitter.bsky.social · May 28

Two multisite projects aimed at scaling up mouse connectomics lost federal funding in recent cuts to Harvard University grants.

By @avaskham.bsky.social

www.thetransmitter.org/funding/mult...

Multisite connectome teams lose federal funding as result of Harvard cuts

The teams aim to develop tools to scale up mouse connectomics in preparation for eventually mapping an entire human brain.

www.thetransmitter.org

3 11 13

deanpospisil.bsky.social @deanpospisil.bsky.social · Apr 11

I am confused. Neither is either, except in a very abstract sense. Is this a counter example?

1

deanpospisil.bsky.social @deanpospisil.bsky.social · Apr 9

All digital twins are models, not all models are digital twins. I like the distinction of a model that predicts a real thing, e.g. a specific neuron in a brain. As opposed to a toy model that predicts qualitative features of a class of neurons in general.

1 1

deanpospisil.bsky.social @deanpospisil.bsky.social · Mar 4

Assuming LDS with synapse count as weight, yes, later…

1

deanpospisil.bsky.social @deanpospisil.bsky.social · Mar 4

In fly its typically 10-100 inputs to a given neuron. In plot below in degree is number of neurons as inputs. Seems pretty doable. Not sure what mouse is like.

1

deanpospisil.bsky.social @deanpospisil.bsky.social · Mar 1

In fly connection probability is 0.00016, I wonder if this goes down with bigger brains b/c geometric constraints?

1 1

deanpospisil.bsky.social @deanpospisil.bsky.social · Feb 28

If you know the non-zero entries of the matrix (connectome) it helps a-lot because vast majority entries are zero: pmc.ncbi.nlm.nih.gov/articles/PMC...
Knowing synapse ‘types’ (from anatomy) could dramatically reduce number of free parameters even more.

From connectome to effectome: learning the causal interaction map of the fly brain

A long-standing goal of neuroscience is to obtain a causal model of the nervous system. This would allow neuroscientists to explain animal behavior in terms of the dynamic interactions between neurons...

pmc.ncbi.nlm.nih.gov

1 3

Reposted

Mengye Ren @mengyer.bsky.social · Dec 7

It’s almost 2025 and the US’s green card backlog for people born in China and India was still 2022 for EB-1 and 2020 for EB-2—visa categories that represent some of the most talented individuals on the planet.

1 1 5

Reposted

Jeremy Berg @jeremymberg.bsky.social · Jan 26

In the present discussions about NIH and the need for reform, two recurring points I am seeing are:

(1) NIH doesn't fund truly high impact research

(2) NIH needs to experiment with alternative approaches to peer review, particularly for high risk-high impact research

1/n

6 60 140

deanpospisil.bsky.social @deanpospisil.bsky.social · Jan 26

Please explain!!!

Reposted

Dave Levitan @davelevitan.bsky.social · Jan 24

More from inside NIH:

Per a source with knowledge, for all internal research (of which there is like $10 billion worth or so), ALL purchasing shut down as of yesterday.

That means gloves, reagents, anything involved with lab work, which means a lot of that work will stop.

88 1.3K 2.6K

deanpospisil.bsky.social @deanpospisil.bsky.social · Jan 22

Why is this considered a problem?

Providing an unavailable dataset e.g., connectome, has led to scientific progress.

Analyzing a model can be more insightful than knowing it is a better fit than a prior model.

deanpospisil.bsky.social @deanpospisil.bsky.social · Jan 16

Eli5 if you have a hot cold neuron but its sometimes wrong, it is helpful to have a neuron that tells you when its wrong.
Statistically var(x+y)=var(x) + var(y) + cov(x,y) so with negative cov you can get less variance (noise).

1

deanpospisil.bsky.social @deanpospisil.bsky.social · Dec 6

Computational models of the brain. Reading from a neurophilosophy undergrad course taught by one of Pat Churchland’s graduate students Brian Keeley. Made the idea of computational models of the brain concrete (and exciting!) for me.

1

deanpospisil.bsky.social @deanpospisil.bsky.social · Nov 29

This is beautiful!

Matt Smear @mattsmear.bsky.social · Nov 25

(I forgot to ask in the thread:)
Please consider turning on the sound for this post. The sniffs & spikes are sonified by converting the event times to MIDI.

Kick drum=inhalation times
Piano=spike times, where each note signifies the spike of a given unit

(code & preprint on the method soon)

Matt Smear @mattsmear.bsky.social · Nov 22

Sniffing helps animals identify smells and connect them to places and events, but noses can’t sense time or place.

How do brains connect odors with internal models of the world?

Our preprint suggests that the olfactory bulb participates in this connection.

www.biorxiv.org/content/10.1...

1