Question on the update for meta learner

[gzliyan113]

Hi, thanks for sharing the code!

I have a question on updating the meta learner: in the original paper, the update for meta learner is interpolation between the previous parameter and the parameters learned by the local learners.
However, it seems that here you are updating the meta learner based on the gradients produced by local learners on the query set. I was wondering is this implementation the same as the original paper's implementation?

Thanks!

[machanic]

@gzliyan113 I think this implementation is not Reptile, but the MAML. As you can see MAML's update is based on the sum of gradients over all tasks.

[tingxueronghua]

I think so, too!

[chenmozxh]

agree with you!

[lzl19971215]

agree, I think this is the implementation of FOMAML.