Improving the Reusability of Pre-trained Language Models in Real-world Applications

TL;DR

This paper introduces Mask-tuning, a training method that improves the generalization and reusability of pre-trained language models on unseen, out-of-distribution data, enhancing their practical application.

Contribution

The paper proposes Mask-tuning, a novel fine-tuning approach that incorporates MLM objectives to boost PLMs' generalization to OOD examples.

Findings

Mask-tuning outperforms existing methods on OOD datasets.

It improves PLMs' performance on in-distribution data.

Enhances the practical reusability of PLMs in real-world scenarios.

Abstract

The reusability of state-of-the-art Pre-trained Language Models (PLMs) is often limited by their generalization problem, where their performance drastically decreases when evaluated on examples that differ from the training dataset, known as Out-of-Distribution (OOD)/unseen examples. This limitation arises from PLMs' reliance on spurious correlations, which work well for frequent example types but not for general examples. To address this issue, we propose a training approach called Mask-tuning, which integrates Masked Language Modeling (MLM) training objectives into the fine-tuning process to enhance PLMs' generalization. Comprehensive experiments demonstrate that Mask-tuning surpasses current state-of-the-art techniques and enhances PLMs' generalization on OOD datasets while improving their performance on in-distribution datasets. The findings suggest that Mask-tuning improves the reusability of PLMs on unseen data, making them more practical and effective for real-world applications.