# On sensitivity of meta-learning to support data Source: [On sensitivity of meta-learning to support data]() Collection index: [llms.txt]() Data updated: 2025-12-24T16:17:15.595Z Authors: Mayank Agarwal, Mikhail Yurochkin, Yuekai Sun Published: 2021-10-26T18:36:37.000Z ## Abstract 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. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Few-shot Image Classification on miniImageNet (test)](): Few-shot Image Classification; miniImageNet (test); Accuracy: not recorded. - [Few-shot Image Classification on FC100 (test)](): Few-shot Image Classification; FC100 (test); Accuracy: not recorded. - [Few-shot Image Classification on CIFAR-FS (test)](): Few-shot Image Classification; CIFAR FS (test); Worst Accuracy: not recorded. ## Original sources and implementation [Original paper]() [PDF]()