/LoRKD

Primary LanguagePython

Low-Rank Knowledge Decomposition for Medical Foundation Models

This repository is an official PyTorch implementation of "Low-Rank Knowledge Decomposition for Medical Foundation Models"

Note: This is a simple version of the code example, more details will be updated soon.

Download Dataset

Run

For example, you can run the following command for LoRKD training.

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python train_lora_shufflenet_medmnist.py

Citation

If you find LoRKD useful for your research or development, please cite the following:

@inproceedings{zhou2024low,
  title={Low-Rank Knowledge Decomposition for Medical Foundation Models},
  author={Zhou, Yuhang and Li, Haolin and Du, Siyuan and Yao, Jiangchao and Zhang, Ya and Wang, Yanfeng},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={11611--11620},
  year={2024}
}