readme.md

Validation Free and Replication Robust Volume-based Data Valuation [NeurIPS'2021]

This repository is the official implementation of the following paper accepted by the Thirty-fifth Conference on Neural Information Processing Systems (NeurIPS) 2021:

Xinyi Xu*, Zhaoxuan Wu*, Chuan Sheng Foo, Bryan Kian Hsiang Low

Validation Free and Replication Robust Volume-based Data Valuation paper

Requirements

To install requirements:

conda env create -f environment.yml

Run synthetic data on baseline distributions

First modify the 'CONFIGS' section in the main.py code, then

mkdir outputs
python main.py

Use real-world datasets

Follow the instructions in the data/ folder.

In the readme.md file under each dataset directory, we specify the collection of dataset files to download and put under the directory.

Plotting the results

Most of the code for plotting figures can be found in the following jupyter notebooks:

• notebooks/Experiments_plots.ipynb
• notebooks/volume_numerical_experiments.ipynb

Other experiments

Code for all other experiments used in the paper can be found unber the following directories:

• robustness experiments
• other experiments

Citing

If you have found our work to be useful in your research, please consider citing it with the following bibtex:

@inproceedings{Xu2021,
 author = {Xu, Xinyi and Wu, Zhaoxuan and Foo, Chuan Sheng and Low, Bryan Kian Hsiang},
 booktitle = {Advances in Neural Information Processing Systems},
 editor = {M. Ranzato and A. Beygelzimer and Y. Dauphin and P.S. Liang and J. Wortman Vaughan},
 pages = {10837--10848},
 publisher = {Curran Associates, Inc.},
 title = {Validation Free and Replication Robust Volume-based Data Valuation},
 volume = {34},
 year = {2021}
}