Better Fine-Tuning by Reducing Representational Collapse
This addresses the problem of unstable fine-tuning for NLP practitioners, offering an incremental improvement over existing trust region approaches.
The paper tackles the instability and representational collapse in fine-tuning pre-trained language models by proposing a simplified trust region method using parametric noise, which matches or exceeds previous methods' performance on tasks like GLUE and generation benchmarks while being faster and reducing collapse.
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.