It is a vanilla python based example following idea of Finn et al Model-Agnostic Meta Learning. In this task, the MAML algorithm is implemented on MNIST dataset. And it aims to achieve better performance on binary classification of two particular digits with knowledge of other digits.
Notice: meta update is achieved by first-order approximation, which essentially reflects gradient according to paper.
python 3
jupyter notebook
- Unzip all files in "/data" folder
- Run python file "src/runner.py" or notebook "src/runner.ipynb"
