85b9a5ac91cd629bd3afe396ec07270a-AuthorFeedback.pdf

R3: Choice of backbone: We have tried to use an architecture (3-4 conv layers + 2 FCs) that is commonly used31 in meta-learning works for its simplicity1 2 3, and have used it to run all our baselines. However, we reiterate that the LLL setting is challenging yet realistic in many cases where it37 isnot feasible tostore allthe within-task data points, and isalso studied inmanyprior works likeAGEM,MER.