Meta-learning algorithms are widely used for few-shot learning. For example, image recognition systems that readily adapt to unseen classes after seeing only a few labeled examples. Despite their success, we show that modern meta-learning algorithms are extremely sensitive to the data used for adaptation, i.e. support data. In particular, we demonstrate the existence of (unaltered, in-distribution, natural) images that, when used for adaptation, yield accuracy as low as 4\% or as high as 95\% on standard few-shot image classification benchmarks. We explain our empirical findings in terms of class margins, which in turn suggests that robust and safe meta-learning requires larger margins than supervised learning.
@article{arxiv.2110.13953,
title = {On sensitivity of meta-learning to support data},
author = {Mayank Agarwal and Mikhail Yurochkin and Yuekai Sun},
journal= {arXiv preprint arXiv:2110.13953},
year = {2021}
}