arXiv Analytics

Sign in

arXiv:2009.07253 [cs.CL]AbstractReferencesReviewsResources

Autoregressive Knowledge Distillation through Imitation Learning

Alexander Lin, Jeremy Wohlwend, Howard Chen, Tao Lei

Published 2020-09-15Version 1

The performance of autoregressive models on natural language generation tasks has dramatically improved due to the adoption of deep, self-attentive architectures. However, these gains have come at the cost of hindering inference speed, making state-of-the-art models cumbersome to deploy in real-world, time-sensitive settings. We develop a compression technique for autoregressive models that is driven by an imitation learning perspective on knowledge distillation. The algorithm is designed to address the exposure bias problem. On prototypical language generation tasks such as translation and summarization, our method consistently outperforms other distillation algorithms, such as sequence-level knowledge distillation. Student models trained with our method attain 1.4 to 4.8 BLEU/ROUGE points higher than those trained from scratch, while increasing inference speed by up to 14 times in comparison to the teacher model.

Related articles: Most relevant | Search more
arXiv:1908.10835 [cs.CL] (Published 2019-08-28)
An Empirical Comparison on Imitation Learning and Reinforcement Learning for Paraphrase Generation
arXiv:1910.13634 [cs.CL] (Published 2019-10-30)
An Augmented Transformer Architecture for Natural Language Generation Tasks
arXiv:2004.14364 [cs.CL] (Published 2020-04-29)
Generating Safe Diversity in NLG via Imitation Learning