Sign in

Martin Hebart

@martinhebart.bsky.social
4.9K followers 607 following 411 posts

Proud dad, Professor of Computational Cognitive Neuroscience, author of The Decoding Toolbox, founder of things-initiative.org our lab 👉 hebartlab.com

PostsRepliesMedia
Reposted by Martin Hebart
Chris Baker @cibaker.bsky.social · 07/10/2026
A new method for translating functional brain responses between species - Spatiotemporal Hyperalignment projects shared information into a common representational space Led by @kbraunlich.bsky.social Marianne Duyck with @bevilconway.bsky.social www.biorxiv.org/content/10.6...
biorxiv.org
Translating functional brain activity between humans and monkeys
Comparing brain responses across species is essential for identifying models of human cognition and testing theories of brain evolution, but differences in brain size, measurement techniques, and sign...
13911
Martin Hebart @martinhebart.bsky.social · 25/09/2026
Together, these results show what structure representations converge on, what factors determine their emergence and how universality could be used of an index of AI-to-human alignment. This could also help develop more human-aligned models in the future.
030
Martin Hebart @martinhebart.bsky.social · 25/09/2026
Universality strongly correlated with human behavioral and monkey brain predictivity. So there is something about universal dimensions that makes them more brain-like and aligned with humans. We could even show causally that aligning models with humans makes their dimensions more universal. 🤯
131
Martin Hebart @martinhebart.bsky.social · 25/09/2026
Universality was not explained by differences in architecture, training data, training objective, model size, or ImageNet performance. However, ...
100
Martin Hebart @martinhebart.bsky.social · 25/09/2026
The results also showed that more universal dimensions were more semantic/conceptual. This was confirmed with additional categorization based analyses, so universality in trained vision models for penultimate layers reflects more high-level content.
130
Martin Hebart @martinhebart.bsky.social · 25/09/2026
We then asked humans to rate dimensions by whether they were visual, semantic, both, or neither (uninterpretable). We found that more universal dimensions were much more interpretable. This means interpretability is an emergent property underlying universality on trained models. 🤯
120
Martin Hebart @martinhebart.bsky.social · 25/09/2026
We first extracted neural network model representations to >22k images and identified sparse non-negative dimensions with a method we call similarity-based representation factorization. We then searched for matching dims between models and assigned them universality scores. arxiv.org/abs/2605.26921
110
Martin Hebart @martinhebart.bsky.social · 25/09/2026
What is the nature of universal representations in AI models, and what determines whether they emerge? Our paper accepted at #neurips2026 led by @florianmahner.bsky.social & @rothj.bsky.social addressed these questions comparing 162 vision models, with intriguing results. arxiv.org/abs/2605.13675 🧵
arxiv.org
Characterizing Universal Object Representations Across Vision Models
Deep neural networks trained with different architectures, objectives, and datasets have been reported to converge on similar visual representations. However, what remains unknown is which visual prop...
34315
Reposted by Martin Hebart
John Timmer @jtimmer.bsky.social · 22/09/2026
Ok, this "there are two separate brains" thing appears to be making the rounds, and I want to be absolutely clear: this is a press office overhyping some perfectly reasonable science that won't come as a surprise to most developmental biologists.
32755238
Martin Hebart @martinhebart.bsky.social · 04/09/2026
Cloud research Connect @cloudresearch.bsky.social is by far the best and better than Mturk ever was.
040
Martin Hebart @martinhebart.bsky.social · 04/09/2026
I think we no longer need what they called artificial artificial AI, and that was part of the business model. I think part of it is bots, and by now they must be really good. But who knows for sure…
000
Martin Hebart @martinhebart.bsky.social · 04/09/2026
Amazon Mechanical Turk (AMT/Mturk) will be shut down as of September 30th. This marks the end of an era. It laid the foundation to the now ubiquitous use of online studies in the psychological and social sciences. It also became too unreliable in recent years. Still, I'm going to miss it.
2113
Reposted by Martin Hebart
Steven Scholte @neurosteven.bsky.social · 31/08/2026
Now out in Current Biology! Object decoding and DNN–brain alignment dissociate: more object information ≠ more alignment. What neural networks share with the brain is mid-level image statistics. Free 50-day access: www.sciencedirect.com/science/arti...
1305
Martin Hebart @martinhebart.bsky.social · 31/08/2026
Two weeks left to sign up for Re:vision! Let's figure out how well visual neuroscience findings replicate! Participants can: - become a coauthor on the paper - win cash prizes up to $2500 - use your replication to write a separate paper. Sign up now, it's super easy: re-vision-initiative.org
01210
Reposted by Martin Hebart
kriegeskorte.bsky.social @kriegeskorte.bsky.social · 30/08/2026
To learn how brains compute, we need to experimentally adjudicate among competing computational hypotheses. How do we do this in the age of complex neural network models? New review paper: “Making models disagree to learn how brains compute” with @talgolanneuro.bsky.social @heikoschuett.bsky.social
1348
Reposted by Martin Hebart
Tal Golan @talgolanneuro.bsky.social · 28/08/2026
How can we design experiments that make computational models disagree? One section of our new @natrevneuro.nature.com Review with @kriegeskorte.bsky.social and @heikoschuett.bsky.social examines studies that used stimulus sets designed to elicit distinct predictions from competing models. 1/16
114156
Reposted by Martin Hebart
Joost de Jong @joost-de-jong.bsky.social · 28/08/2026
🧵 What is the briefest visual response we can elicit? Intuitively, that's the response to the briefest stimulus, aka a flash, right? 📸 No, we can go even briefer! Using a deceptively simple technique, we designed stimuli that elicit briefer-than-brief responses www.biorxiv.org/content/10.6...
biorxiv.org
1135
Martin Hebart @martinhebart.bsky.social · 24/08/2026
P.S.: None of this would have been possible without the special people who were involved in making these datasets reality: Paolo Papale & Pieter Roelfsema for the monkey data TSVD, and the @cibaker.bsky.social lab and my own team for the human data.
000
Martin Hebart @martinhebart.bsky.social · 24/08/2026
Finally, we are now using this approach more broadly to compare representations across domains. Check out Sander's post for more details! Would be curious to hear what you think about it! And if you liked his work, please repost Sander's work and/or my original post! bsky.app/profile/sand...
141
Martin Hebart @martinhebart.bsky.social · 24/08/2026
For example, we found evidence for shared structure related to symbols, including letters and numbers. This may reflect an evolutionary precursor to the human visual representation of symbols already present in monkeys.
240
Martin Hebart @martinhebart.bsky.social · 24/08/2026
What do we learn from this? First, it means we can be more confident about comparing monkey and human results (in IT), while also taking into account where similarities start to break down and differences emerge. Second, we can now use this approach to explore the shared space more comprehensively.
110
Martin Hebart @martinhebart.bsky.social · 24/08/2026
Wait, does this mean there were no differences? Absolutely not! Human fMRI responses preferred human content, monkey incracranial recordings monkey content. Monkeys also leaned more toward animals, humans toward man-made objects. And human IT was more conceptual & macaque IT more visually weighted.
130
Martin Hebart @martinhebart.bsky.social · 24/08/2026
More excitingly, there were many, many other dimensions that monkeys and humans shared. This clearly goes against the idea that monkey IT has only a low-dimensional object representation. It shows a rich, high-dimensional organization that is even shared across species!
130
Martin Hebart @martinhebart.bsky.social · 24/08/2026
What's very cool about his alignment technique is that Sander could inspect the dimensions underlying this alignment. He found the well-known difference between animals and inanimate things, and between spiky and stubby things. But curvature / sharp edges also played a major role!
130
Martin Hebart @martinhebart.bsky.social · 24/08/2026
Sander's comparison across 8640(!) images found strong similarities in how monkeys and humans process objects. He projected both datasets into a common space and found 90(!) shared dimensions! So,at least in inferotemporal cortex humans & monkeys may "see" the world more similarly than we may think!
150
Martin Hebart @martinhebart.bsky.social · 24/08/2026
Previous comparisons like those shown below showed in 92 images that humans and monkeys both represent animals and animal faces differently from inanimate things. But how far does the comparability between species go? Is the alignment restricted to a few dimensions? Or how broad is it really?
Depiction of the seminal study by Kriegeskorte et al. (2008) showing similar representational similarities for 92 animate and inanimate stimuli, with commonalities driven largely by animate-inanimate differences and faces.
110
Martin Hebart @martinhebart.bsky.social · 24/08/2026
Postdoc @sandervanbree.bsky.social tested how far alignment between monkey and human brains can go. He focused on inferotemporal cortex (IT), critical for object processing, using very large monkey recordings and human fMRI with the same 8,640 object images.
An illustration of the shared datasets used for comparison, with two macaques with intracranial recordings in IT and three humans with fMRI recordings. On the top an illustration of the shared object images, in between the two species an illustration of the multivariate cross-species alignment framework.
120
Martin Hebart @martinhebart.bsky.social · 24/08/2026
I have been meaning to write a post about a preprint that I'm pretty excited about: www.biorxiv.org/content/10.6... In neuroscience we often assume that we can learn something general about “the brain” using different species. But is this true? Do monkeys even see the world the way we do?
On the left, an image of a rhesus macaque looking towards its left, and on the right, an AI-generated zoomed-out rendering of a human in the same pose. Both are looking.
36022
Reposted by Martin Hebart
Kurt Fraser @kurtfraser.bsky.social · 19/08/2026
The Department of Psychology at University of Minnesota seeks an Assistant Professor working on computational approaches related to cognition and/or perception (broadly). This would be in my area of the department and I’m happy to answer any questions. Posting : hr.myu.umn.edu/psc/hrprd/EM...
google.com
Careers
Tenure-Track Assistant Professor, Computational Approaches to Perception and/or CognitionDepartment of PsychologyUniversity of Minnesota, College of Liberal ArtsThe Department of Psychology at the University of Minnesota-Twin Cities announces a search for an outstanding scholar to fill a tenure-track assistant professor position focused on computational approaches to perception and/or cognition. This full-time position in the Department’s Cognitive and Brain Sciences (CAB) Training Program will begin in Fall 2027.Applications are invited from researchers who would extend or complement our historic strengths in combining experimental and computational approaches to perception and cognition. Examples of desired research topics include, but are not limited to, computational approaches in human and/or non-human animal models to study perceptual function and sensory loss, learning, motivation, emotion, and/or higher cognition. While we expect the primary appointment to be in our CAB area, there is  potential for cross-affiliation with other departmental areas.The appointment will be 100%-time over the nine-month academic year (late-August to late-May) and will be made at the rank of tenure-track assistant professor, depending on qualifications and experience and consistent with collegiate and University policy. Salary and start-up packages are competitive. The successful applicant should have and is expected to maintain a strong research program, advise graduate students from a variety of different backgrounds, and teach undergraduate and graduate courses, including core courses within the graduate training program. A strong record of teaching and mentorship is desired. The individual will also be expected to contribute to the service needs of the department, college, university, and profession.A successful candidate's application will demonstrate scholarly distinction, a record of publications, teaching commitment and, if applicable to the research program and the candidate’s career stage, evidence of success in securing and/or pursuing extramural funding.Applicants are invited to review the Workload Principles and Guidelines for Regular Faculty in the College of Liberal Arts, as well as the Standards for Promotion and Tenure in the Department of Psychology.
14638
Reposted by Martin Hebart
Sander van Bree @sandervanbree.bsky.social · 12/08/2026
Excited to share our preprint! w/ @martinhebart.bsky.social To understand how primates visually process objects in the world, we rely on both research in human and macaque IT. But what representations of object space are actually shared between them? biorxiv.org/content/10.6... Quick thread 🧵
biorxiv.org
Shared and Distinct High-Dimensional Object Spaces in Human and Macaque Inferotemporal Cortex
Human and macaque studies of inferotemporal cortex (IT) have shaped our understanding of object vision, yet the extent of their representational alignment and the precise nature of this correspondence...
26626
Reposted by Martin Hebart
Russ Poldrack @russpoldrack.org · 11/08/2026
Many of you will have seen the recent post about changes at OSF. If you'd like to learn more about how to use other services (particularly Zenodo) to share data and code, see my book chapter on resource sharing. bettercode-book.org/book-sharing...
bettercode-book.org
12  Sharing Research Objects – Better Code, Better Science
317783
Reposted by Martin Hebart
Lenny van Dyck @levandyck.bsky.social · 05/08/2026
Excited to be at #CCN2026 in New York to present brand-new work with @kathadobs.bsky.social 🥳 By testing hundreds of categories, we found striking differences in functional specialization between visual cortex and DNNs 🧠 Come find me at poster F81 on Thursday 1:45-3:30 p.m. Happy to chat!
1235
Reposted by Martin Hebart
Leyla Isik @lisik.bsky.social · 03/08/2026
Also if you're at #CCN2026 come chat with me about Minds, Machines, and Brains @mmb-journal.bsky.social A new journal we are launching with @mitpress.bsky.social with my co-EICs Kendrick Kay & Ellie Pavlick, and the rest of the wonderful editorial board: direct.mit.edu/mmb/pages/ed...
direct.mit.edu
Editorial Information | Minds, Machines, and Brains | MIT Press
Editorial Information | Minds, Machines, and Brains | MIT Press Editorial Information All inquiries should be directed to mindsmachinesandbrains@gmail.com Co-Editors in Chief Leyla Isik ...
05013
Reposted by Martin Hebart
Freddy Kamps @fkamps.bsky.social · 29/07/2026
New paper! We found basic scene selectivity arises across the scene network early in infancy (i.e., by ~3.5 months old), prior to independent navigation experience www.pnas.org/doi/10.1073/...
pnas.org
The cortical scene processing network emerges in infancy, prior to independent navigation experience | PNAS
Sighted people rely on vision to recognize and navigate the local environment. By adulthood, human cortex contains at least three regions that resp...
27333
Martin Hebart @martinhebart.bsky.social · 02/08/2026
If it does develop them, it will make those of us obsolete who cannot or do not want to keep up with the pace and who cannot do work truly as creative as AI can. Many of us would become research assistants running experiments for others. Either way, it’s going to be wild in the next few years. 3/3
090
Martin Hebart @martinhebart.bsky.social · 02/08/2026
This problem translates to other sciences. If AI doesn’t develop creative ideation outside of formal systems like math and doesn’t develop the ability for building theories, this will lead to a self-limiting process that cuts off the fuel on which science runs. 2/3
1101
Martin Hebart @martinhebart.bsky.social · 02/08/2026
Science isn’t only about solving problems, it’s also about the random process of coming up with new ideas while working on a problem. By solving math problems for us, AI takes away this random element and the theory-building process that happens behind the scenes. 1/3
2489
Martin Hebart @martinhebart.bsky.social · 01/08/2026
Come and join our replication initiative in visual neuroscience! Low bar for entry, ideal for anyone from a Master’s student looking for a thesis to faculty who want to get back to the bench. 🤓 re-vision-initiative.org
re-vision-initiative.org
re:vision | Replication Initiative for Visual Neuroscience
Test whether published visual-neuroscience findings (NSD, THINGS, etc.) generalize to broadly-sampled natural images using the LAION-fMRI (LfMRI) 7T dataset.
02411
Reposted by Martin Hebart
Dota Tianai Dong @dotadotadota.bsky.social · 21/07/2026
1/5 Over a decade of comparing deep neural networks to the human brain—but what have we actually learned? Our new @cp-trendscognsci.bsky.social Feature Review synthesizes a decade of brain–DNN comparisons, asking what they reveal about brain function across vision and language.
14421
Reposted by Martin Hebart
Lenny van Dyck @levandyck.bsky.social · 16/07/2026
Excited to share that our paper is now out in #JNeurosci! We propose a multidimensional framework of high-level visual cortex that reconciles a longstanding debate. Thanks to @kathadobs.bsky.social, @martinhebart.bsky.social, and everyone else for the great discussions along the way. More to come 🧠🌈
doi.org
Multidimensional feature tuning in category-selective areas of human visual cortex
Two prominent accounts describe the functional organization of human high-level visual cortex. A categorical view emphasizes category-selective areas, while a dimensional view highlights continuous fe...
14112
Reposted by Martin Hebart
Zach Nudelman @zednud.bsky.social · 16/07/2026
Excited to share my first #preprint with @matthiasnau.bsky.social on how gaze behavior shapes brain activity during eyes-closed rest and sleep! Gaze-dependent activity was widespread even during sleep, including in many visual cortices, and altered brain-wide functional connectivity! 🧵 1/8
biorxiv.org
Brain-wide gaze-dependent activity during eyes-closed rest and sleep
Gaze behavior and brain activity are tightly coupled during perception and mental imagery. Whether this coupling extends to eyes-closed states remains largely unknown, primarily because measuring eye ...
25116
Martin Hebart @martinhebart.bsky.social · 14/06/2026
Congratulations, Julia!
110
Reposted by Martin Hebart
Mind, Brain and Behavior Research Center - CIMCYC @cimcyc.bsky.social · 09/06/2026
Researchers explore how the brain makes sense of ambiguous images. Even when an image is unclear or incomplete, we often end up recognizing what we are seeing 🧵👇 @martinhebart.bsky.social @lindedomingo.bsky.social @ortiztudela.bsky.social @gonzalezgarcia.bsky.social @jvoeller.bsky.social
cimcyc.ugr.es
How Does the Brain Resolve Ambiguous Images?
Investigadores/as del Centro de Investigación Mente, Cerebro y Comportamiento (CIMCYC) de la Universidad de Granada, en colaboración con Martin N. Hebart (Justus Liebig University Giessen y el Max Pla...
1152
Reposted by Martin Hebart
Michelle Greene @mgreenephd.bsky.social · 02/06/2026
🚨New dataset just dropped🚨 Introducing Places in the Wild: 67,000 RAW-format photographs (45 mpix) densely sampled from 810 places (260 basic-level categories). This is 11x the number of pixels in ImageNet! Preprint is here: arxiv.org/abs/2606.02481 1/
arxiv.org
Places in the Wild: A Large, High-Resolution RAW Photograph Dataset for Ecologically Valid Vision Research
Large image datasets have accelerated progress in cognitive neuroscience and computer vision. However, most datasets are low-resolution, internet-sourced JPEGs with unknown capture conditions and limi...
16632
Martin Hebart @martinhebart.bsky.social · 02/06/2026
They still struggle with shoes though! 😅
110
Martin Hebart @martinhebart.bsky.social · 01/06/2026
Looks like geons are alive and well: this paper was rated among the top 1.8% of all papers at CVPR: bardofcodes.github.io/superfit/
3334
Reposted by Martin Hebart
Philip Sulewski @psulewski.bsky.social · 26/05/2026
Now out in Nature Neuroscience: "Fixation duration on natural scenes is explained by memory encoding not processing demand". www.nature.com/articles/s41... Our eyes don't linger because recognition is hard; they linger to remember. Let me take you on a quick tour. 🧵
nature.com
Fixation duration on natural scenes is explained by memory encoding not processing demand - Nature Neuroscience
By combining magnetoencephalography and eye tracking, this study sheds light on why people fixate on some parts of natural scenes longer than others. Rather than visual complexity, fixation durations ...
323377
Reposted by Martin Hebart
Lauren Atlas @laurenatlas.bsky.social · 20/05/2026
Our multiverse analysis of associations between skin conductance and acute pain is published! >550 participants x 18 SCR pipelines = 1 winning approach to SCR analysis! (Ledalab + artifact detection). thread below. We hope others find this useful! Please RT :) journals.lww.com/pain/fulltex...
02715
Martin Hebart @martinhebart.bsky.social · 27/05/2026
This was a fun collaboration with a great team, and I was happy to play a small part in it!
1191
Martin Hebart @martinhebart.bsky.social · 19/05/2026
P.S.: For more updates, follow @re-vision-init.bsky.social on Bluesky!
000