Interactive Topographic Network modeling framework

Description

Code for training and evaluating Interactive Topographic Network models of visual cortical topography

Citation

@article{Blauch2022Connectivity,
  title={A connectivity-constrained computational account of topographic organization in primate high-level visual cortex},
  author={Nicholas M. Blauch, Marlene Behrmann, David C. Plaut},
  journal={Proceedings of the National Academy of Sciences},
  year={2022},
  issue={199 (3)},
}

Setup

Setup an appropriate virtual environment on a CUDA-ready machine using requirements.txt. with conda:

conda create --name topographic python=3.6.10
conda activate topographic
conda install cudatoolkit=10.1 -c pytorch # if on linux or otherwise needed
pip install -r requirements.txt
# optionally pip install the topographic module 
python setup.py install develop

modify your .bashrc file to set the environment variable SERVER_NAME. and then configure directories as appropriate using toographic/config.py

export SERVERNAME='nick-local'

Configure use of SLURM for job submission if planning to train models (local use is also possible). Other HPC systems are possible but will require separate configuration.
If using SLURM, edit the file run_slurm.sbatch to adjust for your particular setup.
Acquire ImageNet, VGGFace2, and Places365 from common web sources

Pre-train encoders

. scripts/train_encoders.sh

Train ITN models

. scripts/train_all_itns.sh

Data on KiltHub

Alternatively to running the models yourself, you can download all of the pre-trained and fine-tuned models along with relevant results necessary for all plots in the paper on Kilthub: https://doi.org/10.1184/R1/17131319

Plot the results

View the jupyter notebooks; we recommend using Jupyter Lab (which can be installed and launched with:)

conda install -c conda-forge jupyterlab
jupyter-lab

You must have your paths configured and results produced, either through running the experiments yourself, or downloading the KiltHub data and configuring topographic.config.DATA_DIR to point to the directory holding its data.

notebooks/main_model.ipynb shows how to visualize in depth results of the main model and other variants notebooks/variants.ipynb shows how to compare ITN variants, with different architectures and wiring penalties ($\lambda_w$)

questions and bugs:

create an issue. if a bug, please describe in detail what you have tried and any steps necessary to reproduce the issue.

mschrimpf/ITN