Cooperative Distribution Alignment via JSD Upper Bound

Wonwoong Cho*, Ziyu Gong*, David I. Inouye

Purdue University

Neurips 2022

(*Equal contributions)

Source code for the paper:

Cooperative Distribution Alignment via JSD Upper Bound

Code will be updated soon!

Prerequisites

• Linux
• Python 3.8
• CPU or NVIDIA GPU + CUDA CuDNN

Getting Started

Installation

Download every file from https://anonymous.4open.science/r/AUB (Note that the individual file should be downloaded respectively.)

Dataset

The dataset is exactly same with the original MNIST data (http://yann.lecun.com/exdb/mnist/)

Just in case the link above does not work, you can download it here: https://drive.google.com/file/d/1E7Jggb1JCn-D7HazQuWlMxIT69vllFRU/view?usp=drive_link.

unzip data.zip

Environment setup

1. conda env create -f environment.yml
2. source activate aub

Usage

python run.py --multi_gpu False --setting demo --batch_size 128 --gpu_id 0 --lr 2e-4 --lambda_TC 0.0

• The option --multi_gpu is used for GPU parallelization. The code only support for running on a single GPU now. The option --setting is the name of current experiment. The option --batch_size determines how large each batch should be feed to the GPU at once. The value of this option varies among different GPUs. The option --gpu_id select which GPU the experiment will be run on. Default is 0. Learning rate is determined by option --lr. Regularization for AUB is controlled by option lambda_TC, 0 means no regularization.

BibTeX

If you use this code for your research, please cite our paper:

@inproceedings{cho2022AUB,
  title={Cooperative Distribution Alignment via JSD Upper Bound},
  author={Wonwoong Cho and Ziyu Gong and David I. Inouye},
  booktitle={Advances in Neural Information Processing Systems},
  year={2022}
}