Learning Dynamic BERT via Trainable Gate Variables and a Bi-modal Regularizer

Re-assign community