geometric-intel.bsky.social
Research lab at UCSB Engineering revealing the geometric signatures of nature and artificial intelligence | PI: @ninamiolane.bsky.social
7 posts
106 followers
19 following
Prolific Poster
Conversation Starter
comment in response to
post
🔬 We hope that these insights will help neuroscience & AI alike—offering testable predictions for the brain & better memory in AI models.
📖 arxiv.org/abs/2502.17433
@geometric-intel.bsky.social @neurreps.bsky.social @ai-ucsb.bsky.social @ucsbengineering.bsky.social @ucsbece.bsky.social
comment in response to
post
Our study on 35k RNNs reveals that the selection of the mechanism of short-term memory (ie slow-point manifold or limit cycles) depends on:
🤖 the task structure and
🤖 the learning rate of the RNN.
We further derive scaling laws for how long RNNs can store info before failing.
comment in response to
post
Neuroscientists often study sequential neuronal activations—neurons firing one after another—to explain short-term memory.
We show that two geometric mechanisms can drive this:
🌐 Slow-point manifolds,
🌐 Limit cycles.
But which one is used? 🤔
comment in response to
post
Follow our lab's updates @geometric-intel.bsky.social @ai-ucsb.bsky.social @ucsbengineering.bsky.social :
gi.ece.ucsb.edu
comment in response to
post
Want to learn more?🧐
📺 Subscribe to the NeurReps YouTube channel and find more talks by @mweber.bsky.social @kostaspenn.bsky.social @robinwalters.bsky.social @erikjbekkers.bsky.social S. Ravanbakhsh @andyrepair.bsky.social & more!
youtube.com/@neurreps
comment in response to
post
Check out Dr. Poldrack's talk here!
youtu.be/j4_4QSSHugw?...
comment in response to
post
Picture credits: Susana Carmona's Book Cover. Neuromaternal.
comment in response to
post
@ucsantabarbara.bsky.social @uofcalifornia.bsky.social @ucsbece.bsky.social @ucsb-cs.bsky.social @ucsbengineering.bsky.social
comment in response to
post
More @geometric-intel.bsky.social :
gi.ece.ucsb.edu/news/our-pro...
Picture credits: @susanacarmona.bsky.social 's Book Cover. Neuromaternal.
@ai-ucsb.bsky.social @ucsbece.bsky.social @ucsb-cs.bsky.social @uofcalifornia.bsky.social
comment in response to
post
This project was born with the Maternal Brain Project @chrastil.bsky.social @emilyjacobs.bsky.social
nature.com/articles/s41...
We will extend this dataset and leverage AI to predict postpartum conditions like depression.
@cziscience.bsky.social @chanzuckerberg.bsky.social
comment in response to
post
Shout out to my incredible co-authors: Sophia Sanborn (@naturecomputes.bsky.social), Mark Ho (@markkho.bsky.social), Fred Callaway (@fredcallaway.bsky.social), Nathaniel Daw (@nathanieldaw.bsky.social), and Tom Griffiths (@cocoscilab.bsky.social).
comment in response to
post
More info on the workshop. neuroscience.caltech.edu/programs/wor...
comment in response to
post
See our preprint here biorxiv.org/content/10.1... for "Interpretable deep learning for deconvolutional analysis of neural signals".
comment in response to
post
This was a very challenging review to write, but was also a lot of fun due to working with thoughtful co-authors Dylan Martins, Joy Manda, and Phil Parker (@prlparker.bsky.social). Please check it out and let us know what you think.
comment in response to
post
With amazing team @mathildepapillon.bsky.social @adelemyers.bsky.social @franciscoacosta.bsky.social @louisacornelis.bsky.social @abbybertics.bsky.social Daniel Kunin, Fatih Dinc, Simon Mataigne, Guillermo Bernárdez Gil, Sarah Kushner, Luis F. Pereira, Pablo Suarez-Serrato, Alexander West 🌟