Sushrut Thorat

I’m excited to share the first preprint from my PhD project!
Together with Daniel Kaiser (@dkaiserlab.bsky.social), we investigated how internal models shape inter-individual differences in the perception and neural processing of natural scenes.
Preprint: osf.io/preprints/ps...
1/n

The Journal of Neuroscience has updated its policy on supplemental material. https://www.jneurosci.org/content/information-authors#suppl

🚨Super excited that Dartmouth's Society of Fellows is hiring a postdoc with the Program in Cognitive Science 🚨 Specialization in computational and empirical approaches to artificial and natural intelligence, including perception, representation, and complex planning: apply.interfolio.com/176946

Need more fMRI data (beyond the amazing NSD)? Introducing MOSAIC! Incredible effort led expertly by Ben Lahner, with help from grad student Mayukh Deb. Work in collaboration with the amazing Aude Oliva! @neurosky.bsky.social. More below..

0/10 Thanks for the interest in our preprint. Some takes say it negates or fully supports the “manifold hypothesis”, neither quite right. Our results show that if you only focus on the manifold capturing most of task-related variance, you could miss important dynamics that actually drive behavior.

Now only if editors had the same conscience. Coz nowadays proposing a totally new model AND explaining neural data is sometimes not enough it seems 😂

VLMs are truly caught in the middle! We found that they are great at describing what they see AND great at reasoning from text. But, they fail to connect them without an explicit text bridge. Come talk to us about this and more at #Neurips2025

I see you are aware of the Zheng et al. Nat Comms 2024 paper that also builds upon Ali's paper :) (taking it closer to the brain)

Thalamus: a brain-inspired algorithm for biologically-plausible continual learning and disentangled representations Animals thrive in a constantly changing environment and leverage the temporal structure to learn well-factorized causal representations. In contrast, traditional neural networks suffer from forgetting...

Curious what you think. We never made a "release-thread" so I don't expect it would have appeared on most people's radars. It builds upon this groundbreaking paper by Ali Hummos - arxiv.org/abs/2205.11713 - that was *directly* inspired by thalamus-PFC interactions underlying flexible rule inference.

This is a super informative review. I feel everyone might find our recent work interesting in this regard - arxiv.org/abs/2502.15634 - blending ICL/CL to deal with a WCST-like setting, in a neuro-inspired manner! (summary coming soon)

The superior colliculus encodes gaze commands in retinal coordinates - Nature Neuroscience Nature Neuroscience - The superior colliculus encodes gaze commands in retinal coordinates

E.g: when you stimulate anterior / posterior superior colliculus you get fixed-vector / goal-directed saccades respectively. Early reports concluded this means different spatial codes, but later this was explained by a single retina-fixed model with some 3D geometry.
www.nature.com/articles/nn0...

Is there a "sassiness" neuron somewhere I could play with, I wonder.

Got called a "high-agency chaos monkey" by ChatGPT... 🤡

+ Barcelona is an amazing city (one of my favs in EU)! amazing food, and the art scene is crazy good!

I’m all for it but I’ve heard enough complaints of “oh there’s so much out there, how would we ever find the “relevant”, “important” stuff without high impact journals feeding it to us” from senior researchers to wonder if they‘ll buy this. But maybe we don’t need to worry about them 😋

Somehow dissociating between a repository for papers, peer review, and attention might be useful in starting to fix things systemically? Idk.

Something like this?:
Everything -> Preprint -> Society peer-review (centralized by societies - TMLR style rigor check + “a chance to help authors improve their paper”) -> Magazines/Journals solicit (they pay into Society) 1 or 2-pager highlights & Societies solicit reviews periodically.

These latent-space sequence learning methods are turning out to be quite useful in generating brain-aligned representations! We also found an alignment at the level of scene representation with a JEPA-style system - bsky.app/profile/sush...

1/6 New preprint 🚀 How does the cortex learn to represent things and how they move without reconstructing sensory stimuli? We developed a circuit-centric recurrent predictive learning (RPL) model based on JEPAs.
🔗 doi.org/10.1101/2025...
Led by @atenagm.bsky.social @mshalvagal.bsky.social

That’s specifically why I mentioned TMLR and not NeurIPS 😇

Perhaps the “zeitgeist” is right but I definitely don’t feel so. Simplification under a rich umbrella like what TMLR is aiming to do seems totally healthy and fine. Highlights are for X/Bsky/Blogs/Talks/Magazines. Seems to work fine for some of us.

Also having been privy to some of the high-impact publication shenanigans (being in the know, focus on message more than comprehensiveness) my suspicion towards high-IF papers has increased if anything.

I have heard the “there’s too much to read, so we’d like a selective venue” argument before. Unfortunately, in my experience the glam papers have been more about glam than making an important, theory-improving, point (atleast in my sub-fields)

Fair point. Perhaps I really am too young then coz I still don’t perceive the “respect” part wrt journals yet. Papers’ titles (and thereafter the abstracts) have been dictating my pull afaik.

But if journals have field classification, isn’t the issue resolved? We can choose to filter by keywords too? How does the absolute number of papers matter then?

Perhaps I’m quite young in the field and don’t see the value of segregated journals 😋 They anyway have internal classifications with section editors etc.

My point was meant for those that want to do a "minimal change" to their usual practices. If I could have my way, I'd publish everything with TMLR :P (no big neuro equivalents though, perhaps NBDT?)

😁 yup, but Science Advances already publishes cool work in neuro, no? My PhD advisor is one of the editors—he's a hardcore cognitive neuroscientist. I also saw they have cool systems neuro editors too!

Yes!! Also, like me, if you thought Science group of journals is for-profit, you'd be wrong! - www.science.org/doi/10.1126/...

So you CAN publish fully in non-profit journals EVEN IF you want ultra-high IF! Think about it. Time for a change.

a red building on UPENN's campus photographed during the fall

the Philadelphia skyline, with clear skies and autumn trees

starting fall 2026 i'll be an assistant professor at @upenn.edu 🥳

my lab will develop scalable models/theories of human behavior, focused on memory and perception

currently recruiting PhD students in psychology, neuroscience, & computer science!

reach out if you're interested 😊