Cengiz Pehlevan

NEW: #Kempner researchers develop a mean-field theory of task-trained RNNs that bridges random and learned connectivity—and find macaque motor cortex is best captured by an intermediate, task-specific recurrent structure.
Read the blog post 👇
🔗 bit.ly/47f3Ldl

I am totally pumped about this new work . "Task-trained RNNs" are a powerful and influential framework in neuroscience, but have lacked a firm theoretical footing. This work provides one, and makes direct contact with the classical theory of random RNNs:
www.biorxiv.org/content/10.6...

Looking forward to presenting on "How behavior shapes recurrent circuits across sensory systems and species: from vision to touch" at the University of Chicago Neuroscience and ML workshop on Wednesday! Details below 👇🧵

I’m deeply thankful to my supervisor, Mark Brandon (@markbrandonlab.bsky.social) for his patience, guidance, and constant support throughout this project, and to our collaborators in the Cengiz Pehlevan (@cpehlevan.bsky.social) lab at Harvard for their thoughtful and generous contributions.

All theory is wrong until verified by data. Greatly indebted to @mhyaghoubi.bsky.social, @markbrandonlab.bsky.social, @douglasresearch.bsky.social for finding the hippocampus encoding reward prediction! Grateful to my advisor @cpehlevan.bsky.social, @kempnerinstitute.bsky.social.
#RL #hippocampus

Delighted to have contributed to this work. Huge kudos to everyone involved.

I’m very happy to share the latest from my lab published in @Nature

Hippocampal neurons that initially encode reward shift their tuning over the course of days to precede or predict reward.

Full text here:
rdcu.be/eY5nh

Very excited about this new work from the omnipotent Owen, with me and Ashok Litwin-Kumar! Can we reconcile low- and high-dimensional activity in neural circuits by recognizing that these circuits ~multitask~?
(Plausibly, yes 😊)

Schmidt Sciences Awards Early Career Fellowships to Michael Albergo, Melanie Weber - Kempner Institute Two Kempner Institute community members have received AI2050 Fellowships from Schmidt Sciences, a nonprofit organization aimed at accelerating scientific knowledge and breakthroughs. The AI2050 Progra...

Congratulations to #KempnerInstitute community members @msalbergo.bsky.social and @mweber.bsky.social — recipients of @schmidtsciences.bsky.social AI2050 Fellowships! 🎉
Discover their innovative research shaping the future of AI 👉 bit.ly/47Do4R3
#AI

First paper from the lab!
We propose a model that separates estimation of odor concentration and presence and map it on olfactory bulb circuits
Led by @chenjiang01.bsky.social and @mattyizhenghe.bsky.social joint work with @jzv.bsky.social and with @neurovenki.bsky.social @cpehlevan.bsky.social

Connectivity Structure and Dynamics of Nonlinear Recurrent Neural Networks The structure of brain connectivity predicts collective neural activity, with a small number of connectivity features determining activity dimensionality, linking circuit architecture to network-level...

Now in PRX: Theory linking connectivity structure to collective activity in nonlinear RNNs!
For neuro fans: conn. structure can be invisible in single neurons but shape pop. activity
For low-rank RNN fans: a theory of rank=O(N)
For physics fans: fluctuations around DMFT saddle⇒dimension of activity

scipost.org/SciPostPhysL...

Applying to do a postdoc or PhD in theoretical ML or neuroscience this year? Consider joining my group (starting next Fall) at UT Austin!
POD Postdoc: oden.utexas.edu/programs-and... CSEM PhD: oden.utexas.edu/academics/pr...

William Qian, Cengiz Pehlevan: Discovering alternative solutions beyond the simplicity bias in recurrent neural networks https://arxiv.org/abs/2509.21504 https://arxiv.org/pdf/2509.21504 https://arxiv.org/html/2509.21504

⏳ Less than 1 day left until the Brain & Mind Workshop submission deadline!
🔍 Submit to our Finding or Tutorials track on OpenReview.
Findings track submission: openreview.net/group?id=Neu...
Tutorial track submission: openreview.net/group?id=Neu...
More info: data-brain-mind.github.io

Thank you so much for the kind shoutout! Grateful to be part of such a fantastic team.

Frontiers | Summary statistics of learning link changing neural representations to behavior How can we make sense of large-scale recordings of neural activity across learning? Theories of neural network learning with their origins in statistical phy...

Since I'm back on BlueSky - with @frostedblakess.bsky.social and @cpehlevan.bsky.social we wrote a brief perspective on how ideas about summary statistics from the statistical physics of learning could potentially help inform neural data analysis... (1/2)

Convergent motifs of early olfactory processing are recapitulated by layer-wise efficient coding The architecture of early olfactory processing is a striking example of convergent evolution. Typically, a panel of broadly tuned receptors is selectively expressed in sensory neurons (each neuron exp...

Excited to share new computational work, led by @jzv.bsky.social, driven by Juan Carlos Fernandez del Castillo + contribution from Farhad Pashakanloo. We recover 3 core motifs in the olfactory system of evolutionarily distant animals using a biophysically-grounded model + efficient coding ideas!

YouTube video by DARPAtv AIQ: Artificial Intelligence Quantified

Great to have this video about my @darpa.mil Artificial Intelligence Quantified (AIQ) program out! Very exciting program with absolutely fantastic teams. Stay tuned for some jaw dropping announcements!

www.youtube.com/watch?v=KVRF...

I am extremely grateful to be awarded the National University of Singapore (NUS) Development Grant, and to be a Young NUS Fellow! Look forward to collaborating with the Yong Loo Lin School of Medicine on exciting projects. This is my first grant and hopefully many more to come! #NUS #NeuroAI

Simons Foundation Launches Collaboration on the Physics of Learning and Neural Computation Simons Foundation Launches Collaboration on the Physics of Learning and Neural Computation on Simons Foundation

Our new Simons Collaboration on the Physics of Learning and Neural Computation will develop powerful tools from #physics, #math, computer science and theoretical #neuroscience to understand how large neural networks learn, compute, scale, reason and imagine: www.simonsfoundation.org/2025/08/18/s...

Kempner Research Fellowship - Kempner Institute The Kempner brings leading, early-stage postdoctoral scientists to Harvard to work on projects that advance the fundamental understanding of intelligence.

If you work on artificial or natural intelligence and are finishing your PhD, consider applying for a Kempner research fellowship at Harvard:
kempnerinstitute.harvard.edu/kempner-inst...

Simons Foundation Launches Collaboration on the Physics of Learning and Neural Computation Simons Foundation Launches Collaboration on the Physics of Learning and Neural Computation on Simons Foundation

Congratulations to #KempnerInstitute associate faculty member @cpehlevan.bsky.social for joining the new
@simonsfoundation.org Simons Collaboration on the Physics of Learning and Neural Computation!

www.simonsfoundation.org/2025/08/18/s...

#AI #neuroscience #NeuroAI #physics #ANNs

Very excited to lead this new @simonsfoundation.org collaboration on the physics of learning and neural computation to develop powerful tools from physics, math, CS, stats, neuro and more to elucidate the scientific principles underlying AI. See our website for more: www.physicsoflearning.org

🚨 Excited to announce our #NeurIPS2025 Workshop: Data on the Brain & Mind

📣 Call for: Findings (4- or 8-page) + Tutorials tracks

🎙️ Speakers include @dyamins.bsky.social @lauragwilliams.bsky.social @cpehlevan.bsky.social

🌐 Learn more: data-brain-mind.github.io

The post is based on a paper written with Yue M. Lu., @jzv.bsky.social, Anindita Maiti and @cpehlevan.bsky.social evan.

Check it out now at PNAS:

doi.org/10.1073/pnas...

(2/2)

Solvable Model of In-Context Learning Using Linear Attention - Kempner Institute Attention-based architectures are a powerful force in modern AI. In particular, the emergence of in-context learning enables these models to perform tasks far beyond the original next-token prediction...

New in the #DeeperLearningBlog: the #KempnerInstitute's Mary Letey presents work recently published in PNAS that offers generalizable insights into in-context learning (ICL) in an analytically-solvable model architecture.

bit.ly/4lPK15p

#AI @pnas.org

(1/2)

In-context denoising with one-layer transformers: connections between attention and associative memory retrieval We introduce in-context denoising, a task that refines the connection between attention-based architectures and dense associative memory (DAM) networks, also known as modern Hopfield networks. Using a...

At #ICML2025, presenting work done at @flatironinstitute.org w Matt Smart and @albertobietti.bsky.social on in-context denoising (arxiv.org/abs/2502.05164). Come to Matt’s oral, Thursday, 4:15-4:30 PM, West Ballroom A, and see us right after at poster #E-3207, 4:30-7:00 PM, East Exhibition Hall A-B.

Asymptotic theory of in-context learning by linear attention | PNAS Transformers have a remarkable ability to learn and execute tasks based on examples provided within the input itself, without explicit prior traini...

Great to see this one finally out in PNAS! Asymptotic theory of in-context learning by linear attention www.pnas.org/doi/10.1073/... Many thanks to my amazing co-authors Yue Lu, Mary Letey, Jacob Zavatone-Veth @jzv.bsky.social and Anindita Maiti!

#eNeuro: Obeid and Miller identify distinct neural computations in the primary visual cortex that explain how surrounding context suppresses perception of visual figures and features. ‪@harvardseas.bsky.social‬
vist.ly/3n6tfb2