Virtual Data Augmentation: A Robust and General Framework for Fine-tuning Pre-trained Models

Abstract

Recent works have shown that powerful pre-trained language models (PLM) can be fooled by small perturbations or intentional attacks. To solve this issue, various data augmentation techniques are proposed to improve the robustness of PLMs. However, it is still challenging to augment semantically relevant examples with sufficient diversity. In this work, we present Virtual Data Augmentation (VDA), a general framework for robustly fine-tuning PLMs. Based on the original token embeddings, we construct a multinomial mixture for augmenting virtual data embeddings, where a masked language model guarantees the semantic relevance and the Gaussian noise provides the augmentation diversity. Furthermore, a regularized training strategy is proposed to balance the two aspects. Extensive experiments on six datasets show that our approach is able to improve the robustness of PLMs and alleviate the performance degradation under adversarial attacks. Our codes and data are publicly available at \textcolor{blue}{\url{https://github.com/RUCAIBox/VDA}}.

arXiv commentAccepted by EMNLP 2021 (Main Conference)

References in corpus (5)

• FreeLB: Enhanced Adversarial Training for Natural Language Understanding
• Adversarial Robustness through Local Linearization
• Transforming Question Answering Datasets Into Natural Language Inference Datasets
• TAVAT: Token-Aware Virtual Adversarial Training for Language Understanding
• Better Robustness by More Coverage: Adversarial Training with Mixup Augmentation for Robust Fine-tuning