35464c848f410e55a13bb9d78e7fddd0-AuthorFeedback.pdf

RespondtoReviewer1 Acommon biasisthatmeta-learning should tackle transfer learning orfew-shot learning1 problems. However, this is not always the case:the setting of this paperdo not fit nicely with transfer learning or2 few-shotlearning. As pointed out by ICLR 2019 AnonReviewer3 of the6 MAXLpaper,"Moreover,sincethemethodisnotameta-7 learning approachforfew-shot learning,itisnotfairand8 also not appropriate to compare with Prototypical Net-9 work.", Itisnot advisable to evaluate the degree of improvements without con-12 sidering the room available for improvements. Our improvements are13 significant as: (1) they are greater than those achieved by MAXL on14 almostalldatasets(2)according toline240-243, backbones usedinour15 work are already strong, and our work is more effective than naively16 increasingthebackbonedepths. Besides, our model is not sensitive to the choice of datasets.31