CEGA

This is the open-source code for ICML 2025 paper CEGA: A Cost-Effective Approach for Graph-Based Model Extraction and Acquisition.

Setup Environment

conda create --name my_env --file requirements.txt

To run a single experiment, use the command:

python main.py --dataset amazoncomputer --method random

To run ALL attack experiments, use the command:

bash run_bash.sh

The outputs of the experiment are recorded in a timestamped directory under ./output/.

./output/{timestamp}/{dataset}/log/{dataset}/settings.txt

Results can be visualized by using ./plots/model_performance.ipynb. Before running it, update the results folder paths:

datasets = [
    '{timestamp}/{dataset}'
]
cega_results = [
    '{timestamp}/{dataset}'
]