Not to Overfit or Underfit the Source Domains? An Empirical Study of Domain Generalization in Question Answering
This work addresses domain generalization in question answering, offering a novel perspective that could improve model robustness for NLP applications, though it is incremental as it builds on existing techniques like knowledge distillation.
The study challenges the common belief that overfitting to source domains is the main issue in domain generalization, instead showing that underfitting source domains is a primary problem. Experiments on a reading comprehension benchmark reveal that improving source domain learning, such as through knowledge distillation, leads to better zero-shot out-of-domain performance, outperforming existing methods that focus on limiting overfitting.
Machine learning models are prone to overfitting their training (source) domains, which is commonly believed to be the reason why they falter in novel target domains. Here we examine the contrasting view that multi-source domain generalization (DG) is first and foremost a problem of mitigating source domain underfitting: models not adequately learning the signal already present in their multi-domain training data. Experiments on a reading comprehension DG benchmark show that as a model learns its source domains better -- using familiar methods such as knowledge distillation (KD) from a bigger model -- its zero-shot out-of-domain utility improves at an even faster pace. Improved source domain learning also demonstrates superior out-of-domain generalization over three popular existing DG approaches that aim to limit overfitting. Our implementation of KD-based domain generalization is available via PrimeQA at: https://ibm.biz/domain-generalization-with-kd.