Few-shot Adaptation Works with UnpredicTable Data
This work addresses the challenge of scaling few-shot learning for NLP practitioners, but it is incremental as it builds on prior findings about task diversity without fully explaining the underlying mechanisms.
The authors tackled the problem of improving few-shot learning performance by automatically extracting 413,299 tasks from internet tables, which is orders of magnitude larger than existing datasets. Finetuning on this data led to improved performance on NLP tasks, with a narrow subset (software documentation) achieving a mean gain of +7.5% on 52 downstream tasks, beating human-curated datasets.
Prior work on language models (LMs) shows that training on a large number of diverse tasks improves few-shot learning (FSL) performance on new tasks. We take this to the extreme, automatically extracting 413,299 tasks from internet tables - orders of magnitude more than the next-largest public datasets. Finetuning on the resulting dataset leads to improved FSL performance on Natural Language Processing (NLP) tasks, but not proportionally to dataset scale. In fact, we find that narrow subsets of our dataset sometimes outperform more diverse datasets. For example, finetuning on software documentation from support.google.com raises FSL performance by a mean of +7.5% on 52 downstream tasks, which beats training on 40 human-curated NLP datasets (+6.7%). Finetuning on various narrow datasets leads to similar broad improvements across test tasks, suggesting that the gains are not from domain adaptation but adapting to FSL in general. We do not observe clear patterns between the datasets that lead to FSL gains, leaving open questions about why certain data helps with FSL.