Towards Robust Extractive Question Answering Models: Rethinking the Training Methodology

TL;DR

This paper introduces a new training approach for extractive question answering models that enhances robustness to distribution shifts and adversarial attacks without sacrificing in-domain accuracy.

Contribution

It proposes a novel loss function and challenges an implicit assumption in existing datasets, leading to more robust EQA models with improved out-of-domain and adversarial performance.

Findings

Overall F1 score improved by 5.7 points

Enhanced robustness against adversarial attacks

Maintains in-domain performance

Abstract

This paper proposes a novel training method to improve the robustness of Extractive Question Answering (EQA) models. Previous research has shown that existing models, when trained on EQA datasets that include unanswerable questions, demonstrate a significant lack of robustness against distribution shifts and adversarial attacks. Despite this, the inclusion of unanswerable questions in EQA training datasets is essential for ensuring real-world reliability. Our proposed training method includes a novel loss function for the EQA problem and challenges an implicit assumption present in numerous EQA datasets. Models trained with our method maintain in-domain performance while achieving a notable improvement on out-of-domain datasets. This results in an overall F1 score improvement of 5.7 across all testing sets. Furthermore, our models exhibit significantly enhanced robustness against two types of adversarial attacks, with a performance decrease of only about a third compared to the default models.