FairEdit

In this work we introduce FairEdit, the unexplored method of edge addition, accompanied by deletion, to promote fairness. FairEdit performs efficient edge editing by leveraging gradient information of a fairness loss to find edges that improve fairness. We find that FairEdit outperforms standard training for many data sets and GNN methods, while performing comparably to many state-of-the-art methods, demonstrating FairEdit's ability to improve fairness across many domains and models.

Relevent Publication

@misc{loveland2022fairedit,
      title={FairEdit: Preserving Fairness in Graph Neural Networks through Greedy Graph Editing}, 
      author={Donald Loveland and Jiayi Pan and Aaresh Farrokh Bhathena and Yiyang Lu},
      year={2022},
      eprint={2201.03681},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

FairEdit: Preserving Fairness in Graph Neural Networks through Greedy Graph Editing

Reproduce the Results

Requirements

Use environment.yml to build the environment

FairEdit & BruteFoce

Run adjusted_training.py with specified arguments to do traning with various setups. Model weights are saved to weights folder and evaluation metricss are saved to results.

Arguments:

• --model can be ['gcn', 'sage', 'appnp']
• --dataset can be ['german', 'credit', 'bail']
• --training_method can be ['fairedit','brute']
• Hyper-parameters also need to be specified as shown in the example below

Sample code:

python adjusted_training.py --dropout 0.5 --hidden 16 --lr 1e-3 --epochs 1000 --model gcn --training_method fairedit --dataset german --seed 1

Standard Training & Nifty

The same adjusted_traning.py can also be used to run standard training and NIFTY

• Specifically, you just need to change traning method to ['standard', 'nifty']

FairWalk

Code to run FairWalk can be found in models_to_compare_against/FairWalk