Skip to content

xingyousong/ufom

Repository files navigation

UFOM

Code for the paper Debiasing a First-order Heuristic for Approximate Bi-level Optimization at ICML 2021. Note that this is mostly a fork off of the original Reptile codebase, but with a modified reptile.py file, which contains different MAML variants. The UFOM method can be found in the UnbMAML class.

Experiments with hyperparameter tuning configurations can be found in commands.py, while toy experiments can be found in the toy folder.

If you found this codebase to be useful, please consider citing our paper:

@article{ufom,
  author    = {Valerii Likhosherstov and
               Xingyou Song and
               Krzysztof Choromanski and
               Jared Davis and
               Adrian Weller},
  title     = {Debiasing a First-order Heuristic for Approximate Bi-level Optimization},
  journal   = {CoRR},
  volume    = {abs/2106.02487},
  year      = {2021},
  url       = {https://arxiv.org/abs/2106.02487},
  archivePrefix = {arXiv},
  eprint    = {2106.02487}
}

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published