Skip to content

Official implementation of On-Demand Sampling: Learning Optimally from Multiple Distributions (Neurips 2022)

Notifications You must be signed in to change notification settings

ericzhao28/multidistributionlearning

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

On-Demand Sampling: Learning Optimally from Multiple Distributions

This is an early-release of the code used in the experiments of the Neurips 2022 paper On-Demand Sampling: Learning Optimally from Multiple Distributions (HJZ 22).

Instructions

First, download the Waterbirds, MultiNLI, and CelebA datasets to the root of this project. Then, run ready.sh, which will call run.sh. The latter script will run the experiments. When complete, the paper's figures can be reproduced by running the generate_paper_results.py script.

Acknowledgements

This codebase is based in large part on the codebase of the Group DRO implementation of the original authors of S. Sagawa, et al. 2019.

Copyright

THIS SOFTWARE AND/OR DATA WAS DEPOSITED IN THE BAIR OPEN RESEARCH COMMONS REPOSITORY ON OCTOBER 24TH 2022.