Code and results accompanying our paper titled RATT: Leveraging Unlabeled Data to guarantee generalization at ICML 2021 (Long Talk). If you find this repository useful or use this code in your research, please cite the following paper:
Garg, S., Balakrishnan, S., Kolter, Z., Lipton, Z. (2021). RATT: Leveraging Unlabeled Data to Guarantee Generalization. arxiv preprint arXiv:2105.00303.
@inproceedings{garg2021RATT,
title={ {RATT}: Leveraging Unlabeled Data to Guarantee Generalization},
author={Garg, Saurabh and Balakrishnan, Sivaraman and Kolter, Zico and Lipton, Zachary},
year={2021},
booktitle={International Conference on Machine Learning (ICML)}
}
The code is written in Python and uses PyTorch. To install requirements, setup a conda enviornment using the following command:
conda create --file requirements.txt
We simulate the setup for labeled and unlabeled data with the training sets of CIFAR-10, CIFAR-100, MNIST and IMDb. train.py file is the main entry point for training the model and run the code with the following command:
python train.py --lr=0.1 --clean-samples=40000 --noise-samples=10000 --check-steps=100 --wd=0.0005 --dataset=CIFAR10 --arch=Resnet --batch-size=200 --num-classes=10
Change the parameters to your liking and run the experiment. For example, change dataset with varying --dataset and architecture --arch. We implement IMDb dataset with LSTM model in train_imdb_elmo.py.
This repository is licensed under the terms of the Apache-2.0 License.
For more details, refer to the accompanying ICML 2021 paper (Long talk): RATT: Leveraging Unlabeled Data to Guarantee Generalization. If you have questions, please feel free to reach us at sgarg2@andrew.cmu.edu or open an issue.