/csl

Co-Supervised Learning: Improving Weak-to-Strong Generalization with Hierarchical Mixture of Experts

Primary LanguagePythonOtherNOASSERTION

Co-Supervised Learning

This repository is the official implementation of
Co-Supervised Learning: Improving Weak-to-Strong Generalization with Hierarchical Mixture of Experts

The current codebase is a minimalistic version of co-supervised learning, built upon the 'vision' directory of weak-to-strong generalization. It will be continually maintained and updated in the future. If you have any questions or comments, please feel free to raise issues or email yuejiang.liu[at]{epfl.ch,stanford.edu}.

Getting Started

Please refer to the vision directory

Expected Results

Citation

If you find this code useful for your research, please cite the following:

@article{liu2024csl,
  author  = {Yuejiang Liu and Alexandre Alahi},
  title   = {Co-Supervised Learning: Improving Weak-to-Strong Generalization with Hierarchical Mixture of Experts},
  journal = {arXiv preprint 2402.15505},
  year    = {2024},
}

Acknowledgments

  • weak-to-strong repository from OpenAI
  • pre-trained DINOv2 models from Meta AI
  • DomainBed repository from Meta AI