Code and networks for the paper Dubreuil A., Valente A., Beiran M., Mastrogiuseppe F., Ostojic S., "The role of population structure in computations through neural dynamics" Nature Neuroscience, 2022, link, biorXiv link

To install dependencies and use the low_rank_rnns library, simply run in a virtual environment of your choice:

pip install -e .

Cuda is supported but you will need to install the torch package with the appropriate version for your system.

To see how to train a low-rank RNN on a task and fit populations to its parameters, see the tutorial in figures_notebooks/tutorial.ipynb.

Code by Adrian Valente and Alexis Dubreuil.