MATE-KD: Masked Adversarial TExt, a Companion to Knowledge Distillation
This addresses the problem of compressing large pre-trained language models for practical NLP applications, but it is incremental as it builds on existing knowledge distillation and adversarial learning techniques.
The paper tackles improving knowledge distillation for language models by introducing MATE-KD, a text-based adversarial training algorithm that perturbs text to maximize divergence between teacher and student logits, resulting in a 6-layer RoBERTa-based model outperforming BERT-Large on the GLUE benchmark.
The advent of large pre-trained language models has given rise to rapid progress in the field of Natural Language Processing (NLP). While the performance of these models on standard benchmarks has scaled with size, compression techniques such as knowledge distillation have been key in making them practical. We present, MATE-KD, a novel text-based adversarial training algorithm which improves the performance of knowledge distillation. MATE-KD first trains a masked language model based generator to perturb text by maximizing the divergence between teacher and student logits. Then using knowledge distillation a student is trained on both the original and the perturbed training samples. We evaluate our algorithm, using BERT-based models, on the GLUE benchmark and demonstrate that MATE-KD outperforms competitive adversarial learning and data augmentation baselines. On the GLUE test set our 6 layer RoBERTa based model outperforms BERT-Large.