On the Convergence of the Shapley Value in Parametric Bayesian Learning Games [ICML-2022]

Official implementation of our ICML'22 paper "On the Convergence of the Shapley Value in Parametric Bayesian Learning Games" (21.9% acceptance rate).

Requirements

1. Linux machine (experiments were run on Ubuntu 18.04.5 LTS and Ubuntu 20.04.2 LTS)
2. Anaconda (alternatively, you may install the packages in environment.yml manually)

Setup

1. Run the following command to install the required Python packages into a new environment named PBLG using Anaconda.

conda env create -f environment.yml

Running experiments

In the main directory,

1. Change current environment to the PBLG environment.

conda activate PBLG

2. Run the desired experiment. Files with names exp_*.py are scripts for different experiments.

• For instance, exp_CaliH-P1.py runs the experiment on the California housing dataset by varying the hyperparameters w.r.t. Player 1 (P1).
• Similarly, exp_CaliH-multi.py runs the experiment on the California housing dataset with multiple players (i.e., 4).

License

This code is released under the MIT License.

Citing our paper

If you find our paper relevant or use our code in your research, please consider citing our paper:

@InProceedings{Agussurja2022,
  title={Incentivizing collaboration in machine learning via synthetic data rewards},
  author={Lucas Agussurja and Xinyi Xu and Bryan Kian Hsiang Low},
  booktitle={Proc. ICML},
  year={2022}
}