A Graph-Theoretic Framework for Understanding Open-World Semi-Supervised Learning

Yiyou Sun, Zhenmei Shi, Yixuan (Sharon) Li

This repo contains the reference source code in PyTorch of the SORL framework. For more details please check our paper A Graph-Theoretic Framework for Understanding Open-World Semi-Supervised Learning (NeurIPS 23, spotlight).

Dependencies

The code is built with the following libraries:

• PyTorch==1.7.1

Usage

Get Started

• Download models pre-trained by unsupervised spectral contrastive loss here and put under the pretrained folder.

• To train and evaluate on CIFAR-100/10, run

./run.sh

Citing

If you find our code useful, please consider citing:

@inproceedings{
    sun2023sorl,
    title={A Graph-Theoretic Framework for Understanding Open-World Semi-Supervised Learning},
    author={Yiyou Sun and Zhenmei Shi and Yixuan Li},
    booktitle={NeurIPS},
    year={2023},
    url={https://openreview.net/forum?id=ZITOHWeAy7}
}