Fine-Tuning Pre-Trained Language Models Effectively by Optimizing Subnetworks Adaptively

Large-scale pre-trained language models have achieved impressive results on a wide range of downstream tasks recently.