Timezone: »

 
Poster
On sensitivity of meta-learning to support data
Mayank Agarwal · Mikhail Yurochkin · Yuekai Sun

Tue Dec 07 08:30 AM -- 10:00 AM (PST) @
in Poster Session 1 »

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.

Keywords: Machine Learning · Robustness · Vision · Meta Learning · Few Shot Learning

Author Information

Mayank Agarwal (IBM Research AI, MIT-IBM Watson AI Lab)
Mikhail Yurochkin (IBM Research, MIT-IBM Watson AI Lab)
Yuekai Sun (University of Michigan)

More from the Same Authors