Foundation models in neuroscience predict brain activity at unprecedented accuracy. But prediction ≠ understanding, and we should avoid conflating the two.
New essay now out:
For 200 years, Weber's law was the most replicated finding in psychophysics but nobody could explain why it holds. Turns out the answer was hiding in the data everyone ignored — not how often you get it right, but how long it takes you to decide.
The Chomsky hierarchy isn't a linguistic curiosity — it's a map of what kinds of memory architecture different classes of computation require. Do the boundaries between computational classes correspond to boundaries in neural architecture?
New Field Note: a Deep dive into the work of our guest lecturers this week. Three papers from the Dynamical Inference lab on contrastive learning for neural time series—CEBRA, DCL, xCEBRA. What identifiability actually means, when you need dynamics modeling, applications and implications.
New Journal Club: Neural manifolds are maturing from visualization trick to biological claim. But if population activity lives on low-dimensional manifolds, what constrains the geometry?
Every computational model needs numbers. But most experimental data lives in figures, not files. On the hidden labor of turning papers into models, and why open data is infrastructure for theory.
Over the past week, I’ve been involved in organizing an Advanced Course on Systems and Computational Neuroscience. Offering modern perspectives via guest lectures and extensive practical tutorials, we...
open.substack.com
The population doctrine—the view that populations, not individual neurons, constitute the fundamental unit of computation—has been gaining ground for years.
Foundation models in neuroscience: representational alignment versus mechanistic understanding
open.substack.com
On the hidden labor of turning papers into models, and why open data is infrastructure for theory
open.substack.com
When I published last week’s data-driven prelude to CoSyNe 2026, I promised to report from the frontlines and intended to write daily dispatches with fresh perspectives and an almost real-time pulse o...
Wilson–Cowan. FitzHugh–Nagumo. Brunel. Kuramoto.
The detailed models that motivated those reductions are mostly forgotten outside their original communities. The reductions are still being taught.
A field note on why every mechanistic model deserves a small theoretical cousin.
Renato Duarte
Most of the modeling work I do aims at maintaining a high degree of biological fidelity while elucidating computational primitives.
I tracked every keyword in 22 years of Cosyne abstracts to map how computational neuroscience evolved — from Bayesian brains to neural manifolds to LLMs — and where it's heading next.
open.substack.com
Tracking the intellectual DNA of Computational and Systems Neuroscience through its flagship meeting
