FreeLB: Enhanced Adversarial Training for Natural Language Understanding

TL;DR

FreeLB introduces a novel adversarial training method that enhances language model robustness and accuracy by adding perturbations to embeddings, leading to improved performance on multiple NLP benchmarks.

Contribution

The paper proposes FreeLB, a new adversarial training algorithm that increases invariance in embeddings and improves natural language understanding models' performance.

Findings

Improves BERT-base test score from 78.3 to 79.4 on GLUE

Achieves state-of-the-art 85.44% on ARC-Easy

Boosts RoBERTa-large performance on CommonsenseQA

Abstract

Adversarial training, which minimizes the maximal risk for label-preserving input perturbations, has proved to be effective for improving the generalization of language models. In this work, we propose a novel adversarial training algorithm, FreeLB, that promotes higher invariance in the embedding space, by adding adversarial perturbations to word embeddings and minimizing the resultant adversarial risk inside different regions around input samples. To validate the effectiveness of the proposed approach, we apply it to Transformer-based models for natural language understanding and commonsense reasoning tasks. Experiments on the GLUE benchmark show that when applied only to the finetuning stage, it is able to improve the overall test scores of BERT-base model from 78.3 to 79.4, and RoBERTa-large model from 88.5 to 88.8. In addition, the proposed approach achieves state-of-the-art single-model test accuracies of 85.44\% and 67.75\% on ARC-Easy and ARC-Challenge. Experiments on CommonsenseQA benchmark further demonstrate that FreeLB can be generalized and boost the performance of RoBERTa-large model on other tasks as well. Code is available at \url{https://github.com/zhuchen03/FreeLB .