Efficient neural codes naturally emerge through gradient descent learning

This repository hosts code for "Efficient neural codes naturally emerge through gradient descent learning" available at https://www.nature.com/articles/s41467-022-35659-7

This is a collaboration with the Stocker lab.

Local setup

conda env create -f environment.yml

Organization

All subpanels for figures in the paper can be produced by running code in the appropriate notebook in figure_notebooks. Many figures can be run locally, but some will strictly require cuda. All notebbooks can be run on Google Colab.

External files

Imagenet crops can be downloaded at: https://drive.usercontent.google.com/download?id=1mF46SUDKzG0LkWkNGV1fP2hTEgW5WbF\_

	Run	View
Figure 2
Figure 3
Figure 4
Figure 6
Train on Rotated CIFAR