NeuralTaskonomy

Inferring the Similarity of Task-Derived Representations from Brain Activity

This repository contains code for the Neural Taskonomy paper, accepted to NeurIPS 2019.

Setup and Installation

Step 1: Clone the code from Github

git clone https://github.com/ariaaay/NeuralTaskonomy.git
cd NeuralTaskonomy

You will also need to clone the taskonomy model bank.

git clone https://github.com/StanfordVL/taskonomy/tree/master/taskbank

BOLD5000 data and stimuli is available for download here.

Step 2: Install requirements

Requirements.txt contains the necessary package for to run the code in this project.

python3 -m venv venv
source venv/bin/activate
pip install -r requirement.txt --no-index

Please also follow installation page to install another environment to use taskonomy model bank.

Step 3: Generate model activations from each task specific models for all BOLD5000 Images

sh scripts/generate_taskonomy_features.sh

To run encoding models (ridge regression) using task presentaitons from all tasks on ROI data and whole brain data.

sh scripts/taskrepr_ROI_train.sh
sh scripts/taskrepr_wholebrain.sh

To run permutatation tests on ROI and whole brain data

sh scripts/taskrepr_ROI_permutation.sh
sh scripts/taskrepr_wholebrain_permutation.sh

This permutates the brain response 5000 times to obtain the null distribution.

To process permutation results:

cd code
python process_permuation_results.py --subj $subj
python make_task_network.py --use_mask_corr --subj $subj --empirical
python run_significance_test.py --subj $subj --whole_brain --use_empirical_p

To generate task similarity matrix:

python make_task_matrix.py --method "cosine" --use_mask_corr --empirical

To generate task trees:

python make_task_tree.py --subj $subj --method masked_corr