MLTI

About

Source code of the paper Meta-Learning with Fewer Tasks through Task Interpolation

If you find this repository useful in your research, please cite the following paper:

@inproceedings{yao2022meta,
  title={Meta-Learning with Fewer Tasks through Task Interpolation},
  author={Yao, Huaxiu and Zhang, Linjun and Finn, Chelsea},
  booktitle={Proceeding of the 10th International Conference on Learning Representations},
  year={2022} 
}

Data

We have put the related datasets in Google Drive

Usage

Dependence

• python 3.*
• Pytorch 1.10+

Details

1. Bio folder is for NCI and Metabolism
2. Image_NLS is for miniImagenet-S, ISIC and DermNet-S
3. Please check run.sh in each folder for running scripts

Code of MLTI with ProtoNet is released. Code of MLTI with MAML coming soon.