Description

This repository accompanies the preprint "Mahuas, G., Buffet, T., Marre, O., Ferrari, U., & Mora, T. (2024). Strong, but not weak, noise correlations are beneficial for population coding." Read the preprint on arXiv.

It contains the code necessary to reproduce the results shown in Fig. 4, illustrating the effect of noise correlations on stimulus encoding at different spatial scales.

Manual

To reproduce panels 4B and 4C from precomputed data, run the notebook located at ./analyze/plot_panels.ipynb.

To redo the analysis from scratch, follow these steps in order:

1. Run simulate/simulate.sh to generate the synthetic datasets.
2. Run train_decoder/train_decoder.sh to train the decoders.
3. Run analyze/analyze_decoder.sh to compute the mutual informations and the noise synergy from the decoders.
4. Plot the results using analyze/plot_panels.ipynb.

The Bash scripts need to be made executable by running: chmod +x <script_name>.sh.