| layout | default |
|---|---|
| title | About Us |
| desc | Our mission and vision |
| nav | About |
We're a group of theoretical and computational neuroscientists based at the Center for Neural Science at NYU and the Center for Computational Neuroscience at the Flatiron Institute.
How do you build biological intelligence? Biological intelligence is as wide-ranging and diverse as the problems presented by survival and the species that navigate them. It is not one faculty but many, from the vocal mimicry of songbirds to the social dynamics of primates. Its speed, adaptability, and energy efficiency remain unrivaled by any engineered system.
Our goal is to understand how the ingredients of biological computation, particularly those neglected in current neural network models, come together to produce a system vastly more powerful than these constituent parts. How is neural computation organized across different species, and are there generalizable principles that can be used to describe it?
While we work on a variety of problems and systems in theoretical neuroscience, our approach in each case built on a few key ideas:
Our lab takes a "bottom up" approach to theory. We typically start from data, devising new data analysis methods where necessary, and move toward generalizable principles.
Nothing in neuroscience makes sense except in light of behavior.1 We prefer behaviors like vocal learning and grasping and stimuli like movies because they give us the opportunity to study the brain in something closer to its normal working mode.
Modern artificial neural networks begin as clean slates and learn exclusively from data. Real organisms, on the other hand, are the result of developmental programs and environmental interactions that see them capable of incredible behaviors at birth. Many of the most impressive examples of intelligent behavior are the result of brains that do not begin from scratch but are born ready to learn.
We code in the open. We share data. Communicating science requires finding and telling the stories in our data, but these stories are worthless if they don't stand up to scrutiny from the community.
Almost all our projects are done in close collaboration with the experimentalists who generate the data we model. Our code and algorithms are designed to solve real scientific problems faced by real users.
Footnotes
-
With apologies to Theodosius Dobzhansky. ↩