UDApter -- Efficient Domain Adaptation Using Adapters

TL;DR

This paper introduces two adapter-based methods for efficient unsupervised domain adaptation in NLP, significantly reducing parameter count while maintaining or improving performance on tasks like NLI and sentiment classification.

Contribution

It presents novel adapter-based approaches that enable parameter-efficient domain adaptation, outperforming some existing methods and requiring only a small fraction of model parameters.

Findings

Outperforms strong baselines in NLI and sentiment tasks.

Achieves comparable or better results than unsupervised domain adaptation methods.

Uses only a small fraction of parameters compared to full fine-tuning.

Abstract

We propose two methods to make unsupervised domain adaptation (UDA) more parameter efficient using adapters, small bottleneck layers interspersed with every layer of the large-scale pre-trained language model (PLM). The first method deconstructs UDA into a two-step process: first by adding a domain adapter to learn domain-invariant information and then by adding a task adapter that uses domain-invariant information to learn task representations in the source domain. The second method jointly learns a supervised classifier while reducing the divergence measure. Compared to strong baselines, our simple methods perform well in natural language inference (MNLI) and the cross-domain sentiment classification task. We even outperform unsupervised domain adaptation methods such as DANN and DSN in sentiment classification, and we are within 0.85% F1 for natural language inference task, by fine-tuning only a fraction of the full model parameters. We release our code at https://github.com/declare-lab/domadapter