Improving Question Answering Performance Using Knowledge Distillation and Active Learning

TL;DR

This paper introduces a combined knowledge distillation and active learning approach to significantly reduce the complexity and data requirements of question answering systems, achieving comparable performance with fewer resources.

Contribution

It presents a novel KD method for compressing BERT and integrates AL strategies to minimize annotation efforts, enabling high performance with less data and smaller models.

Findings

Model achieves performance of 6-layer TinyBERT and DistilBERT with only 2% of parameters.

State-of-the-art results on SQuAD with just 20% of training data.

Reduces computational and annotation costs significantly.

Abstract

Contemporary question answering (QA) systems, including transformer-based architectures, suffer from increasing computational and model complexity which render them inefficient for real-world applications with limited resources. Further, training or even fine-tuning such models requires a vast amount of labeled data which is often not available for the task at hand. In this manuscript, we conduct a comprehensive analysis of the mentioned challenges and introduce suitable countermeasures. We propose a novel knowledge distillation (KD) approach to reduce the parameter and model complexity of a pre-trained BERT system and utilize multiple active learning (AL) strategies for immense reduction in annotation efforts. In particular, we demonstrate that our model achieves the performance of a 6-layer TinyBERT and DistilBERT, whilst using only 2% of their total parameters. Finally, by the integration of our AL approaches into the BERT framework, we show that state-of-the-art results on the SQuAD dataset can be achieved when we only use 20% of the training data.