The Trade-offs of Domain Adaptation for Neural Language Models
It provides theoretical insights into domain adaptation trade-offs for language models, which is incremental but clarifies existing practices.
This work analyzes how the benefit of training language models on out-of-domain versus in-domain data depends on dataset sizes and distribution distances, showing that out-of-domain pre-training followed by in-domain fine-tuning achieves better generalization than either approach alone.
This work connects language model adaptation with concepts of machine learning theory. We consider a training setup with a large out-of-domain set and a small in-domain set. We derive how the benefit of training a model on either set depends on the size of the sets and the distance between their underlying distributions. We analyze how out-of-domain pre-training before in-domain fine-tuning achieves better generalization than either solution independently. Finally, we present how adaptation techniques based on data selection, such as importance sampling, intelligent data selection and influence functions, can be presented in a common framework which highlights their similarity and also their subtle differences.