LearningMotorFeedback

A biologically plausible generative network to distinguish self- from externally generated optic flow patterns.

This repository provides the code for the experiments in our paper Learning to segment self-generated from externally caused optic flow through sensorimotor mismatch circuits (Brucklacher*, Pezzulo, Mannella, Galati and Pennartz 2023). Please find the full paper at this link and cite when using the code: https://www.biorxiv.org/content/10.1101/2023.11.15.567170v1

* Corresponding author. Email: [email protected]

Installation

• Conda

• Setup the conda environment motorpred_env by running:

  conda env create -f environment.yml

• With the activated environment, manually run:

  pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu118

• With the activated environment, install the local package 'mspc' to allow absolute imports of modules. To do so run the following from directory 'LearningMotorFeedback':

  pip install -e .

Recreate figures from the paper

• Recreate figures for the microcircuit by running the respective fig<figure_number>.py files located in /model1_global/

• Recreate figures for the retinotopic model by running /model2_retinotopic/create_figures/recreate_all.py