On Measuring the Intrinsic Few-Shot Hardness of Datasets

• Main Authors: Zhao, Xinran, Murty, Shikhar, Manning, Christopher D
• Format: Journal Article
• Language: English
• Published: 16.11.2022
• Subjects: Computer Science - Computation and Language
• DOI: 10.48550/arxiv.2211.09113

While advances in pre-training have led to dramatic improvements in few-shot learning of NLP tasks, there is limited understanding of what drives successful few-shot adaptation in datasets. In particular, given a new dataset and a pre-trained model, what properties of the dataset make it \emph{few-shot learnable} and are these properties independent of the specific adaptation techniques used? We consider an extensive set of recent few-shot learning methods, and show that their performance across a large number of datasets is highly correlated, showing that few-shot hardness may be intrinsic to datasets, for a given pre-trained model. To estimate intrinsic few-shot hardness, we then propose a simple and lightweight metric called "Spread" that captures the intuition that few-shot learning is made possible by exploiting feature-space invariances between training and test samples. Our metric better accounts for few-shot hardness compared to existing notions of hardness, and is ~8-100x faster to compute.