Sign in

Raj Magesh

@raj-magesh.org
74 followers 157 following 33 posts

Cognitive computational neuroscience | Postdoc at Pitt advised by Marlene Behrmann | Currently working on a few fun sEEG projects :D raj-magesh.org

PostsRepliesMedia
Raj Magesh @raj-magesh.org · 15/07/2026
A (related) lovely short story: terrybisson.com/theyre-made-... And a modern twist: maxleiter.com/blog/weights
terrybisson.com
They’re Made Out of Meat - TERRY BISSON of the UNIVERSE
Copyright, Terry Bisson, 1991 Originally published in OMNI, 1991, and featured in HARPER’S and around the internet since. It has even made its way into several books on consciousness and brain science...
021
Raj Magesh @raj-magesh.org · 10/07/2026
Very curious about this: could you provide a reference? (PS: love your videos, keep making them!)
010
Raj Magesh @raj-magesh.org · 18/06/2026
Much respect, and thank you! 🫡 (Also consider using pandoc to convert it to HTML and publishing on a website)
010
Raj Magesh @raj-magesh.org · 18/06/2026
Are these notes publicly available/accessible, by any chance?
100
Raj Magesh @raj-magesh.org · 03/06/2026
I'd go one step further: train on GPL code and CC-BY-SA data? You should be forced to release your weights! Distilling the open commons to sell it back to the public as a closed-source product feels to me like such a disservice to all of us.
120
Reposted by Raj Magesh
Thomas Yeo @bttyeo.bsky.social · 22/05/2026
Here's bonus slides on cross-validation tests, separate from our preprint. Covering: 1. paired (sign-flip) permutation test 2. label-swap permutation test 3. sample-level vs fold-averaged stats 4. a common misapplication of the corrected t-test 5. three bootstrap variants 1/N
13015
Raj Magesh @raj-magesh.org · 23/05/2026
Very cool: it's amazing that this doesn't rely on paired data! @mickbonner.bsky.social and I recently found that the shared geometry is characterized by a ~universal power-law: doi.org/10.1371/jour..., Fig 2 Now I'm curious if we could have done it without the paired stimuli...
doi.org
Universal scale-free representations in human visual cortex
Author summary The human cerebral cortex is thought to encode sensory information in population activity patterns, but the statistical structure of these population codes has yet to be characterized. ...
131
Reposted by Raj Magesh
Martin Hebart @martinhebart.bsky.social · 17/05/2026
I'm proud to say we are releasing LAION-fMRI, a densely sampled 7T fMRI dataset of natural images, with very broad stimulus sampling for testing countless hypotheses and for deeply exploring brain representations. The dataset is now available at laion-fmri.hebartlab.com What does LAION-fMRI offer? 🧵
laion-fmri.hebartlab.com
LAION-fMRI - a 7T fMRI dataset of human vision
LAION-fMRI (LfMRI / LAION MRI dataset): 5 subjects, 25,052 launch-release natural images, 165 acquired 7T fMRI sessions with single-trial GLMsingle betas, retinotopy, localizers, and diffusion.
312857
Raj Magesh @raj-magesh.org · 15/05/2026
If only the caduceus were replaced by a rod of Asclepius... en.wikipedia.org/wiki/Rod_of_...
en.wikipedia.org
Rod of Asclepius - Wikipedia
020
Reposted by Raj Magesh
Sepiedeh Keshavarzi @sepikeshavarzi.bsky.social · 07/04/2026
When universities and research institutions become military targets, academics cannot remain silent. I co-initiated an open letter calling for global academic solidarity and support for affected students and scholars. Please share. to read and sign: sites.google.com/view/protect...
03317
Reposted by Raj Magesh
Minas Karamanis @minaskar.bsky.social · 30/03/2026
Hey, I wrote a thing about AI in astrophysics ergosphere.blog/posts/the-ma...
ergosphere.blog
The machines are fine. I'm worried about us.
On AI agents, grunt work, and the part of science that isn't replaceable.
1101780530
Reposted by Raj Magesh
Roselyne Chauvin @roselynechauvin.bsky.social · 13/03/2026
The 6x US memory champion – Nelson Dellis – can memorize a deck of cards in 40 seconds and knows the first 10K digits of pi. To figure out how, he let us peak inside his brain. Here is what we learned in our precision brain mapping study www.biorxiv.org/content/10.6... youtube.com/shorts/MryMq...
youtube.com
how does his brain do it ? #neuroscience #memory #sport Nelson Dellis 6x US memory champion
YouTube video by Roselyne Chauvin
58432
Reposted by Raj Magesh
Manasi Malik @manasimalik.bsky.social · 26/02/2026
Excited to share new work on how the brain makes social inferences from visual input! 🧠👯‍♂️ (With @lisik.bsky.social , @shariliu.bsky.social, @tianminshu.bsky.social , and Minjae Kim!) www.biorxiv.org/content/10.6...
15217
Reposted by Raj Magesh
Tomás Ryan @tjryan.bsky.social · 12/02/2026
Do you work or study in the fields of psychology, neuroscience, computer science, artificial intelligence, or philosophy? What does the term 'representation' mean to you? We invite you to participate in a brief survey on key conceptual questions across fields. eu.surveymonkey.com/r/VX9GNXM
46660
Raj Magesh @raj-magesh.org · 05/02/2026
New typography idea: replace consecutive dots with a bar Even worse idea: if the dots are close enough within a word, join them with an arc
110
Reposted by Raj Magesh
Kelsey Han @chihyehan.bsky.social · 30/01/2026
Human visual cortex representations may be much higher-dimensional than earlier work suggested, but are these higher dimensions of cortical activity actually relevant to behavior? Our new paper tackles this by studying how different people experience the same movies. 🧵 www.cell.com/current-biol...
cell.com
High-dimensional structure underlying individual differences in naturalistic visual experience
Han and Bonner reveal that individual visual experience arises from high-dimensional neural geometry distributed across multiple representational scales. By characterizing the full dimensional spectru...
26016
Raj Magesh @raj-magesh.org · 18/12/2025
Yeah, definitely! An relevant paper along these lines is www.nature.com/articles/nat..., where they show dimensionality collapse on error trials in monkey PFC representations!
nature.com
The importance of mixed selectivity in complex cognitive tasks - Nature
When an animal is performing a cognitive task, individual neurons in the prefrontal cortex show a mixture of responses that is often difficult to decipher and interpret; here new computational methods...
000
Raj Magesh @raj-magesh.org · 18/12/2025
Sorry, I'd missed this sub-thread! Yes, several prior reports of low-D representations were because of deliberate constraints to measure behavioral relevance. Here, we only consider cross-trial/cross-subject reliability, not task-related constraints (a very interesting Q in its own right).
000
Raj Magesh @raj-magesh.org · 17/12/2025
Also: the ease of reaching out to the devs who *actually wrote* the software and getting timely responses from them. And how easy it is to contribute bugfixes. Even if the frequency of bugs is higher, the total annoyance is much lower, perhaps because I feel like I have agency. Long live FOSS!
010
Raj Magesh @raj-magesh.org · 15/12/2025
I think more the latter than the former. But my point is simpler: I think neuroscience experiments often yield low-D manifolds because of simplicity in inputs (e.g. carefully controlled stimuli) and easy tasks. I expect naturalistic stimuli and behaviors would elicit more high-D representations.
110
Reposted by Raj Magesh
Mick Bonner @mickbonner.bsky.social · 15/12/2025
Prediction: task-based optimization will ultimately prove to have a relatively minor role in DNN models of the ventral stream. Although tasks (including self-supervised ones) are currently crucial, there are signs that a simpler approach is possible. A thread:
1209
Raj Magesh @raj-magesh.org · 12/12/2025
I agree that relative measures are cleaner to measure and easier to interpret! Our point in this paper is mainly that the absolute dimensionality is much higher than previously thought throughout visual cortex! And so we might need different approaches to understand these high-D data.
030
Reposted by Raj Magesh
UniReps @unireps.bsky.social · 12/12/2025
📢The UniReps x @ellis.eu speaker series is back! Come join us in our next appointment 18th December 4 pm CET with @meenakshikhosla.bsky.social and Raj Magesh Gauthaman🔵🔴
0126
Reposted by Raj Magesh
Mick Bonner @mickbonner.bsky.social · 12/12/2025
Hopkins Cog Sci is hiring! We have two open faculty positions: one in vision, and one language. Please repost!
03134
Raj Magesh @raj-magesh.org · 12/12/2025
Yeah, given all the limitations, it's amazing how there's still so much stimulus-related information in BOLD signals! In Fig S12 (journals.plos.org/ploscompbiol...) we find power-law spectra in a monkey electrophysiology dataset too. And the same in mouse Ca-imaging: www.nature.com/articles/s41...
journals.plos.org
110
Raj Magesh @raj-magesh.org · 12/12/2025
But also, we're binning the eigenspectrum heavily to measure this small-but-nonzero signal in the tail! This is a tradeoff: we lose spectral resolution but at least we can measure the signal there.
020
Raj Magesh @raj-magesh.org · 12/12/2025
The nice thing about the estimator we're using in the paper is that if there is no stimulus-related signal (i.e. generalizes across repeated presentations and new stimuli), the expected value of the variance is 0. So what we're seeing significantly above zero is not noise.
110
Raj Magesh @raj-magesh.org · 12/12/2025
Ohhh I see what you meant! I've been using "high" and "low" variance to refer to the first few dimensions and the tail of the eigenspectrum respectively. Yeah, in principle, noise should definitely inflate the tail of the eigenspectrum (also the rest, but less noticeably).
110
Raj Magesh @raj-magesh.org · 12/12/2025
Thanks! The cross-decomposition method we're using measures variance that generalizes (i) across multiple presentations of the stimuli and (ii) to a held-out test set, so I'm not too worried about that---we are measuring only stimulus-related signal. (I think you meant low variance?)
110
Raj Magesh @raj-magesh.org · 12/12/2025
Yep, I think many tasks often used in neuroscience won't require attention to many features, but actual naturalistic behavior is probably way more high-dimensional. www.pnas.org/doi/full/10....
pnas.org
PNAS
Proceedings of the National Academy of Sciences (PNAS), a peer reviewed journal of the National Academy of Sciences (NAS) - an authoritative source of high-impact, original research that broadly spans...
110
Raj Magesh @raj-magesh.org · 12/12/2025
I'll refactor it into a standalone tool at some point when I get the time. 🙃 But the sklearn implementation is likely sufficient for most purposes.
010
Raj Magesh @raj-magesh.org · 12/12/2025
I think the best place to start would be this implementation of cross-decomposition in sklearn: scikit-learn.org/stable/modul... I've written a GPU-accelerated version that does other stuff too (permutation tests, etc.) but it's unfortunately not quite plug-and-play (github.com/BonnerLab/sc...).
scikit-learn.org
PLSSVD
100
Raj Magesh @raj-magesh.org · 12/12/2025
Yep, at some point in the process, the relevant info must be extracted for task purposes, and a low-D manifold is what I'd expect to see there. Though it seems that throughout visual cortex at least, the code remains pretty high-dimensional (though how much ends up being used on a task is unclear).
110
Raj Magesh @raj-magesh.org · 12/12/2025
Also, while I think many would agree visual representations are high-dimensional, often our datasets and tools have been too limited to detect it. Estimates of visual cortex dimensionality have traditionally been much lower (~10s-100), not the unbounded power-law we're reporting here.
110
Raj Magesh @raj-magesh.org · 12/12/2025
I tend to think of these representations as being a rich, general-purpose feature bank that can be easily read out from for a variety of tasks. But yeah, I'm sure different latent subspaces are differentially activated based on task demands.
010
Raj Magesh @raj-magesh.org · 12/12/2025
Yeah, that's an important point! Our analysis here only measures reliability of the representation across trials/held-out stimuli, not whether the info is used for downstream processing. I'm also curious how dimensionality depends on task demands, but that's hard to answer with this dataset.
210
Raj Magesh @raj-magesh.org · 12/12/2025
But also, networks do have pretty high-dimensional representations in general, often with power-law statistics too! A nice example is in proceedings.neurips.cc/paper_files/...
proceedings.neurips.cc
$\alpha$-ReQ : Assessing Representation Quality in Self-Supervised Learning by measuring eigenspectrum decay
020
Raj Magesh @raj-magesh.org · 12/12/2025
Yeah, compression of info is something that often happens close to the final layers of DNNs, likely because networks are often trained on a more limited task than an open-ended system like our brains. e.g. networks trained on CIFAR-10 often end up lower-dimensional than those trained on CIFAR-100
030
Raj Magesh @raj-magesh.org · 12/12/2025
Yeah, there are definitely analogous findings in DNNs! I particularly like Figure 7 in arxiv.org/abs/2204.06125 as an example of high-dimensional representations being useful in DNNs.
arxiv.org
Hierarchical Text-Conditional Image Generation with CLIP Latents
Contrastive models like CLIP have been shown to learn robust representations of images that capture both semantics and style. To leverage these representations for image generation, we propose a two-s...
130
Raj Magesh @raj-magesh.org · 12/12/2025
(Apologies, I'm not sure if I'm threading correctly; I'm splitting up a single response into multiple comments due to the incredibly low character limit (!?!))
000
Raj Magesh @raj-magesh.org · 12/12/2025
There was some variation in the size of V1 across participants but that shouldn't affect our results beyond having less data to estimate dimensionality when there are fewer voxels. I'm not quite sure what you meant about V4; could you elaborate or point me to a paper?
100
Raj Magesh @raj-magesh.org · 12/12/2025
We pooled voxels across both hemispheres to have better statistical power (but results were qualitatively similar in each hemisphere separately too).
100
Raj Magesh @raj-magesh.org · 12/12/2025
The regions we analyzed were defined in the original NSD data paper (www.nature.com/articles/s41...) where V1 through hV4 were manually drawn based on the results of a separate population receptive field mapping experiment.
nature.com
A massive 7T fMRI dataset to bridge cognitive neuroscience and artificial intelligence - Nature Neuroscience
The authors measured high-resolution fMRI activity from eight individuals who saw and memorized thousands of annotated natural images over 1 year. This massive dataset enables new paths of inquiry in ...
200
Raj Magesh @raj-magesh.org · 11/12/2025
Yes! We ran the same analysis on some monkey electrophysiology data (the THINGS ventral stream spiking dataset) and observed a similar power-law spectrum of reliable stimulus-related variance shared between macaques (S12 Fig in the paper: journals.plos.org/ploscompbiol...).
journals.plos.org
130
Reposted by Raj Magesh
Mick Bonner @mickbonner.bsky.social · 11/12/2025
Dimensionality reduction may be the wrong approach to understanding neural representations. Our new paper shows that across human visual cortex, dimensionality is unbounded and scales with dataset size—we show this across nearly four orders of magnitude. journals.plos.org/ploscompbiol...
722165