UDALM: Unsupervised Domain Adaptation through Language Modeling

TL;DR

This paper presents UDALM, a novel fine-tuning method for pretrained language models that enhances unsupervised domain adaptation by combining classification and masked language modeling losses, leading to improved performance.

Contribution

UDALM introduces a mixed loss fine-tuning procedure for better domain adaptation of language models, demonstrating robustness and efficiency across multiple domain pairs.

Findings

Achieves 91.74% accuracy on Amazon Reviews dataset.

Performance scales with target data amount.

Effective as a stopping criterion during training.

Abstract

In this work we explore Unsupervised Domain Adaptation (UDA) of pretrained language models for downstream tasks. We introduce UDALM, a fine-tuning procedure, using a mixed classification and Masked Language Model loss, that can adapt to the target domain distribution in a robust and sample efficient manner. Our experiments show that performance of models trained with the mixed loss scales with the amount of available target data and the mixed loss can be effectively used as a stopping criterion during UDA training. Furthermore, we discuss the relationship between A-distance and the target error and explore some limitations of the Domain Adversarial Training approach. Our method is evaluated on twelve domain pairs of the Amazon Reviews Sentiment dataset, yielding $91.74\%$ accuracy, which is an $1.11\%$ absolute improvement over the state-of-the-art.