DE$^3$-BERT: Distance-Enhanced Early Exiting for BERT based on Prototypical Networks

TL;DR

DE$^3$-BERT introduces a novel early exiting method that combines local entropy and global distance information using prototypical networks, improving inference speed and accuracy trade-offs in BERT models.

Contribution

This paper proposes DE$^3$-BERT, a new framework that leverages class prototypes and distance metrics to enhance early exiting decisions in BERT, addressing limitations of local-only approaches.

Findings

Outperforms state-of-the-art models on GLUE benchmark

Achieves better speed-accuracy trade-offs with minimal overhead

Validates effectiveness across different speed-up ratios

Abstract

Early exiting has demonstrated its effectiveness in accelerating the inference of pre-trained language models like BERT by dynamically adjusting the number of layers executed. However, most existing early exiting methods only consider local information from an individual test sample to determine their exiting indicators, failing to leverage the global information offered by sample population. This leads to suboptimal estimation of prediction correctness, resulting in erroneous exiting decisions. To bridge the gap, we explore the necessity of effectively combining both local and global information to ensure reliable early exiting during inference. Purposefully, we leverage prototypical networks to learn class prototypes and devise a distance metric between samples and class prototypes. This enables us to utilize global information for estimating the correctness of early predictions. On this basis, we propose a novel Distance-Enhanced Early Exiting framework for BERT (DE$^3$-BERT). DE$^3$-BERT implements a hybrid exiting strategy that supplements classic entropy-based local information with distance-based global information to enhance the estimation of prediction correctness for more reliable early exiting decisions. Extensive experiments on the GLUE benchmark demonstrate that DE$^3$-BERT consistently outperforms state-of-the-art models under different speed-up ratios with minimal storage or computational overhead, yielding a better trade-off between model performance and inference efficiency. Additionally, an in-depth analysis further validates the generality and interpretability of our method.