Model-Agnostic Meta-Learning on Omniglot, miniImageNet, and CIFAR-FS

This repo extends the original MAML code (link, ICML 2017 paper) [1] with CIFAR-FS (CIFAR few shot) classification based on the splits in Luca et al. [2]. It includes code for running the few-shot supervised learning domain experiments, including sinusoid regression, Omniglot classification, MiniImagenet classification, and the newly added CIFAR-FS classification.

For the experiments in the RL domain, see this codebase.

A report detailing the implementation details and reproducibility can be found here.

Dependencies

This code requires the following:

• python 2.* or python 3.*

• TensorFlow v1.0+

These results were achieved using TensorFlow-gpu 1.8.0, CUDA 9.0 and cudnn 7.

Data

For the Omniglot, MiniImagenet and CIFAR-FS data, see the usage instructions in data/omniglot_resized/resize_images.py and data/miniImagenet/proc_images.py and data/CIFARFS/get_cifarfs.py respectively.

Usage

To run the code, see the usage instructions at the top of main.py.

Results

After 60,000 iterations, with a 95% confidence interval, and 10 finetune steps:

Cite this work

If you use (part of) this code or work, please cite the following work:

@article{devosreproducing,
  title={Reproducing Meta-learning with differentiable closed-form solvers},
  author={Devos, Arnout and Chatel, Sylvain and Grossglauser, Matthias},
  booktitle={International Conference on Learning Representations},
  year={2019},
  url={https://openreview.net/forum?id=BJx0N2I6IN},
}

References

[1] Finn, Chelsea, Pieter Abbeel, and Sergey Levine. "Model-agnostic meta-learning for fast adaptation of deep networks." arXiv preprint arXiv:1703.03400 (2017).

[2] Bertinetto, Luca, et al. "Meta-learning with differentiable closed-form solvers." arXiv preprint arXiv:1805.08136 (2018).