I have a short question regarding the results.
In the case of 5way-5shot classification (2 modes), you report 73% accuracy for Multi-MAML. But actually MAML trained on 5way-5shot yields 99.9% accuracy for Omniglot and 63% for MiniImageNet (see MAML paper or any other paper that reimplemented MAML). So actually Multi-MAML should yield above 80% accuracy. How did you evaluate it to get to 73% ?
Thanks in advance.
I have a short question regarding the results.
In the case of 5way-5shot classification (2 modes), you report 73% accuracy for Multi-MAML. But actually MAML trained on 5way-5shot yields 99.9% accuracy for Omniglot and 63% for MiniImageNet (see MAML paper or any other paper that reimplemented MAML). So actually Multi-MAML should yield above 80% accuracy. How did you evaluate it to get to 73% ?
Thanks in advance.