SpikeBERT: A Language Spikformer Learned from BERT with Knowledge Distillation

TL;DR

SpikeBERT introduces a deep spiking Transformer for language tasks, trained via a two-stage knowledge distillation from BERT, achieving competitive accuracy with lower energy consumption.

Contribution

The paper develops SpikeBERT, a novel deep spiking Transformer model for language understanding, trained with a two-stage knowledge distillation from BERT, enabling efficient and effective language processing.

Findings

SpikeBERT outperforms existing SNNs on text classification.

SpikeBERT achieves comparable results to BERT with less energy.

Two-stage knowledge distillation improves SNN training for language tasks.

Abstract

Spiking neural networks (SNNs) offer a promising avenue to implement deep neural networks in a more energy-efficient way. However, the network architectures of existing SNNs for language tasks are still simplistic and relatively shallow, and deep architectures have not been fully explored, resulting in a significant performance gap compared to mainstream transformer-based networks such as BERT. To this end, we improve a recently-proposed spiking Transformer (i.e., Spikformer) to make it possible to process language tasks and propose a two-stage knowledge distillation method for training it, which combines pre-training by distilling knowledge from BERT with a large collection of unlabelled texts and fine-tuning with task-specific instances via knowledge distillation again from the BERT fine-tuned on the same training examples. Through extensive experimentation, we show that the models trained with our method, named SpikeBERT, outperform state-of-the-art SNNs and even achieve comparable results to BERTs on text classification tasks for both English and Chinese with much less energy consumption. Our code is available at https://github.com/Lvchangze/SpikeBERT.