Domain-agnostic Question-Answering with Adversarial Training

TL;DR

This paper introduces an adversarial training approach for domain-agnostic question answering, enabling models to generalize across domains without fine-tuning, demonstrated on the MRQA Shared Task 2019.

Contribution

It proposes a novel adversarial training framework combining a QA model and discriminator for domain-invariant feature learning in QA tasks.

Findings

Outperforms baseline models on MRQA Shared Task 2019

Demonstrates improved domain generalization in QA

Validates effectiveness of adversarial training in domain adaptation

Abstract

Adapting models to new domain without finetuning is a challenging problem in deep learning. In this paper, we utilize an adversarial training framework for domain generalization in Question Answering (QA) task. Our model consists of a conventional QA model and a discriminator. The training is performed in the adversarial manner, where the two models constantly compete, so that QA model can learn domain-invariant features. We apply this approach in MRQA Shared Task 2019 and show better performance compared to the baseline model.