Armen Aghajanyan, Akshat Shrivastava, Anchit Gupta, Naman Goyal, Luke Zettlemoyer, Sonal Gupta
Although widely adopted, existing approaches for fine-tuning pre-trained language models have been shown to be unstable across hyper-parameter settings, motivating recent work on trust region methods. In this paper, we present a simplified and efficient method rooted in trust region theory that replaces previously used adversarial objectives with parametric noise (sampling from either a normal or uniform distribution), thereby discouraging representation change during fine-tuning when possible without hurting performance. We also introduce a new analysis to motivate the use of trust region methods more generally, by studying representational collapse; the degradation of generalizable representations from pre-trained models as they are fine-tuned for a specific end task. Extensive experiments show that our fine-tuning method matches or exceeds the performance of previous trust region methods on a range of understanding and generation tasks (including DailyMail/CNN, Gigaword, Reddit TIFU, and the GLUE benchmark), while also being much faster. We also show that it is less prone to representation collapse; the pre-trained models maintain more generalizable representations every time they are fine-tuned.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Text Summarization | Reddit TIFU | ROUGE-1 | 30.31 | BART+R3F |
| Text Summarization | Reddit TIFU | ROUGE-2 | 10.98 | BART+R3F |
| Text Summarization | Reddit TIFU | ROUGE-L | 24.74 | BART+R3F |
| Text Summarization | GigaWord | ROUGE-1 | 40.45 | BART-RXF |
| Text Summarization | GigaWord | ROUGE-2 | 20.69 | BART-RXF |
| Text Summarization | GigaWord | ROUGE-L | 36.56 | BART-RXF |
| Text Summarization | CNN / Daily Mail | ROUGE-1 | 44.38 | BART+R3F |
| Text Summarization | CNN / Daily Mail | ROUGE-2 | 21.53 | BART+R3F |
| Text Summarization | CNN / Daily Mail | ROUGE-L | 41.17 | BART+R3F |
| Abstractive Text Summarization | CNN / Daily Mail | ROUGE-1 | 44.38 | BART+R3F |
| Abstractive Text Summarization | CNN / Daily Mail | ROUGE-2 | 21.53 | BART+R3F |
| Abstractive Text Summarization | CNN / Daily Mail | ROUGE-L | 41.17 | BART+R3F |