Meta-learning training process

[shellin-star]

hi，thanks to your great works！I have been curious how the meta learning mentioned in the article is implemented in the code.
I have read it several times, and it feels like this is a regular gan training process rather than MAML.
Because I don't know much about meta-learning, can you give me some suggestions？

thank you again!

[jychoi118]

In my opinion, this work is more close to model-based meta-learning rather than optimization-based meta-learning like MAML. Model-based meta-learning is predicting model parameters with neural network (meta-learner) according to few-data. In this case, the embedder predicts parameters of AdaIN layers.
Meta-learning can be devided into 3 methods: model-based, optimization-based, and metric-based.