How Effective is Task-Agnostic Data Augmentation for Pretrained Transformers?

Task-agnostic forms of data augmentation have proven widely effective in\ncomputer vision, even on pretrained models. In NLP similar results are reported\nmost commonly for low data regimes, non-pretrained models, or situationally for\npretrained models. In this paper we ask how effective these techniques really\nare when applied to pretrained transformers. Using two popular varieties of\ntask-agnostic data augmentation (not tailored to any particular task), Easy\nData Augmentation (Wei and Zou, 2019) and Back-Translation (Sennrichet al.,\n2015), we conduct a systematic examination of their effects across 5\nclassification tasks, 6 datasets, and 3 variants of modern pretrained\ntransformers, including BERT, XLNet, and RoBERTa. We observe a negative result,\nfinding that techniques which previously reported strong improvements for\nnon-pretrained models fail to consistently improve performance for pretrained\ntransformers, even when training data is limited. We hope this empirical\nanalysis helps inform practitioners where data augmentation techniques may\nconfer improvements.\n

Paper

Similar papers

© 2026 NYSGPT2525 LLC