Distilling Task-Specific Knowledge from BERT into Simple Neural Networks

TL;DR

This paper shows that simple neural networks can be made competitive for NLP tasks by distilling knowledge from complex models like BERT, achieving similar performance with much fewer parameters and faster inference.

Contribution

The authors introduce a knowledge distillation method from BERT into lightweight neural networks, enabling competitive NLP performance without complex architectures or additional data.

Findings

Comparable results to ELMo on multiple NLP tasks

Achieved 100x fewer parameters and 15x faster inference

Effective knowledge transfer from BERT to simple models

Abstract

In the natural language processing literature, neural networks are becoming increasingly deeper and complex. The recent poster child of this trend is the deep language representation model, which includes BERT, ELMo, and GPT. These developments have led to the conviction that previous-generation, shallower neural networks for language understanding are obsolete. In this paper, however, we demonstrate that rudimentary, lightweight neural networks can still be made competitive without architecture changes, external training data, or additional input features. We propose to distill knowledge from BERT, a state-of-the-art language representation model, into a single-layer BiLSTM, as well as its siamese counterpart for sentence-pair tasks. Across multiple datasets in paraphrasing, natural language inference, and sentiment classification, we achieve comparable results with ELMo, while using roughly 100 times fewer parameters and 15 times less inference time.