Marius Pachitariu
@marius10p.bsky.social
1.8K followers 670 following 22 posts
Lab head @HHMIJanelia neuroscience, deep learning, large-scale recordings #Kilosort, #Suite2p, #Cellpose, #Rastermap, #Facemap
Posts Media Videos Starter Packs
Reposted by Marius Pachitariu
computingnature.bsky.social
Is visual cortex more like chatGPT or a masked autoencoder? Come to my BCCN seminar tomorrow at 10am ET on zoom (or read the paper) to find out! #neuroscience

Zoom: urldefense.com/v3/__https:/...
Paper: www.biorxiv.org/content/10.1...
qqzhang.bsky.social
Does predictive coding work in SPACE or in TIME? Most neuroscientists assume TIME, i.e. neurons predict their future sensory inputs. We show that in visual cortex predictive coding actually works across SPACE, just like the original Rao+Ballard theory #neuroscience
www.biorxiv.org/cgi/content/...
marius10p.bsky.social
Yes! More generally for affective and/or very salient outcomes, everything could be different. We only measured visual cortex in benign "daily life" scenarios.
marius10p.bsky.social
Another way to think of it is novel items/situations versus novel temporal order of familiar items. However, it could be different in non-sensory brain areas, or in other species, or in other modalities.
marius10p.bsky.social
New work from the lab, mismatch responses consistent with spatial but not temporal predictive coding theory.
qqzhang.bsky.social
Does predictive coding work in SPACE or in TIME? Most neuroscientists assume TIME, i.e. neurons predict their future sensory inputs. We show that in visual cortex predictive coding actually works across SPACE, just like the original Rao+Ballard theory #neuroscience
www.biorxiv.org/cgi/content/...
Reposted by Marius Pachitariu
qqzhang.bsky.social
Does predictive coding work in SPACE or in TIME? Most neuroscientists assume TIME, i.e. neurons predict their future sensory inputs. We show that in visual cortex predictive coding actually works across SPACE, just like the original Rao+Ballard theory #neuroscience
www.biorxiv.org/cgi/content/...
marius10p.bsky.social
TetO‐jGCaMP8s is a drop in replacement for the @crisniell.bsky.social mice with all the nice properties of 8s. High, dense expression, and doesn't have the aberrant activity. Was used here: www.nature.com/articles/s41...
Reposted by Marius Pachitariu
hhmijanelia.bsky.social
💻A new class of simplified ‘minimodels’ developed by Fengtong Du, Miguel Angel Núñez-Ochoa, @marius10p.bsky.social & @computingnature.bsky.social can predict the visual responses of individual neurons in the primary visual cortex, helping shed light on how we see.👀
➡️ www.janelia.org/news/new-com...
marius10p.bsky.social
Your brain learns for you even when you're not actively learning. Maybe. (in mice).
hhmijanelia.bsky.social
🚶Aimlessly wandering around a city or exploring the new mall may seem unproductive, but new research led by @zh0ng0.bsky.social @marius10p.bsky.social @computingnature.bsky.social suggests it could play an important role in how our brains learn.‬‬ 🧠
🔗 www.janelia.org/news/zoning-...
Reposted by Marius Pachitariu
gabrielmstine.bsky.social
Our paper on NHP neuropixels is finally out in Nat Neuro! These probes have already been transformative and will usher in a new era of primate neuroscience. I am extremely proud to have played a very small role in this project. I can't wait to see what our community discovers. tinyurl.com/54u3hrj8
Large-scale high-density brain-wide neural recording in nonhuman primates - Nature Neuroscience
Neuropixels 1.0 NHP is a 45-mm, high-density silicon probe capable of recording large numbers of neurons with single-neuron resolution from most areas in a macaque’s brain.
tinyurl.com
Reposted by Marius Pachitariu
nmwilkinson.bsky.social
AI doesn't have to be shit
marius10p.bsky.social
🚀🔬🦠 Releasing 🤖Cellpose-SAM🤖, a cellular segmentation algorithm with superhuman generalization 🦸‍♀️. Try it now on 🤗 huggingface.co/spaces/mouse...

paper: www.biorxiv.org/content/10.1...
w/ @computingnature.bsky.social 1/n
Reposted by Marius Pachitariu
hhmijanelia.bsky.social
🚶Aimlessly wandering around a city or exploring the new mall may seem unproductive, but new research led by @zh0ng0.bsky.social @marius10p.bsky.social @computingnature.bsky.social suggests it could play an important role in how our brains learn.‬‬ 🧠
🔗 www.janelia.org/news/zoning-...
Reposted by Marius Pachitariu
helenkmatthews.bsky.social
Everyone in the lab is raving about Cellpose-SAM. Segmenting the previously unsegmentable across a range of different cell types and markers.
marius10p.bsky.social
🚀🔬🦠 Releasing 🤖Cellpose-SAM🤖, a cellular segmentation algorithm with superhuman generalization 🦸‍♀️. Try it now on 🤗 huggingface.co/spaces/mouse...

paper: www.biorxiv.org/content/10.1...
w/ @computingnature.bsky.social 1/n
Reposted by Marius Pachitariu
michalwj.bsky.social
Wouldn't it be great if we could not only image large connectomic volumes but also completely reconstruct them? And if a whole mouse brain project didn't cost billions?

With the PATHFINDER preprint (www.biorxiv.org/content/10.1...), we preview a future where it doesn't have to.
Reposted by Marius Pachitariu
princetonupress.bsky.social
A neuroscientist’s bold proposal for tackling one of the greatest challenges of our time—brain and mental illnesses.

Nicole C. Rust's Elusive Cures publishes June 10. Learn more and preorder yours: press.princeton.edu/books/hardco... (5 August UK pub) @nicolecrust.bsky.social
A neuroscientist's bold proposal for tackling one of the greatest challenges of our time--brain and mental illnesses
Reposted by Marius Pachitariu
mameister4.bsky.social
If you're planning a course with mathematical methods content, or use such methods in your own work, please take a look at "Mathematics in Biology", by Meister, Lee, and Portugues, published at MIT Press. 1/2 @portugueslab.bsky.social mitpress.mit.edu/978026204940...
Mathematics in Biology
Biology has turned into a quantitative science. The core problems in the life sciences today involve complex systems that require mathematical expression, ye...
mitpress.mit.edu
marius10p.bsky.social
Ok, yes, this one will be tough no matter what. It might work a little better if there was more context, for example if the FOV was larger.
marius10p.bsky.social
Could you share an image please?
Reposted by Marius Pachitariu
carlosbrody.bsky.social
we're crowd-sourcing a searchable repository of tangible benefits stemming from federally-funded research. Come enjoy the great stories; or send in an idea; or volunteer to join the team.

publicusaresearchbenefits.com

please share and re-share so we get more great stories in there!
Searchable database of tangible benefits that federally-funded research gave us.
A crowd-sourced site. Health and Well-being. National Security. Prosperity.
publicusaresearchbenefits.com
marius10p.bsky.social
I don't think we have good examples to look at this, but you can test it directly on the website or in the GUI and let us know.
marius10p.bsky.social
Cellpose-SAM is currently available in the GUI.
marius10p.bsky.social
The Cellpose-SAM backbone, yes. The SAM encoder weights were re-trained on a lot of cellular data to predict the Cellpose flows.
marius10p.bsky.social
The model can be finetuned for other tasks such as panoptic segmentation with relatively little effort, comparing favorably to the best methods.
Panoptic segmentation challenge with four classes of ROIs and Cellpose-SAM performance compared to the challenge winners before and after t he challenge.