Recall and Learn: Fine-tuning Deep Pretrained Language Models with Less Forgetting

TLDR

Deep pretrained language models excel when pretraining precedes fine‑tuning, yet this sequential paradigm often suffers catastrophic forgetting, limiting performance. The authors propose a recall‑and‑learn mechanism to reduce forgetting during fine‑tuning by jointly learning pretraining and downstream tasks. The framework employs a Pretraining Simulation module that recalls pretraining knowledge without data and an Objective Shifting module that gradually emphasizes downstream objectives, all integrated into the RecAdam optimizer. Experiments demonstrate state‑of‑the‑art GLUE results and show that BERT‑base with the proposed method surpasses directly fine‑tuned BERT‑large on average performance.

Abstract

Deep pretrained language models have achieved great success in the way of pretraining first and then fine-tuning. But such a sequential transfer learning paradigm often confronts the catastrophic forgetting problem and leads to sub-optimal performance. To fine-tune with less forgetting, we propose a recall and learn mechanism, which adopts the idea of multi-task learning and jointly learns pretraining tasks and downstream tasks. Specifically, we introduce a Pretraining Simulation mechanism to recall the knowledge from pretraining tasks without data, and an Objective Shifting mechanism to focus the learning on downstream tasks gradually. Experiments show that our method achieves state-of-the-art performance on the GLUE benchmark. Our method also enables BERT-base to achieve better average performance than directly fine-tuning of BERT-large. Further, we provide the open-source RecAdam optimizer, which integrates the proposed mechanisms into Adam optimizer, to facility the NLP community.