PADA: Example-based Prompt Learning for on-the-fly Adaptation to Unseen Domains

Abstract Natural Language Processing algorithms have made incredible progress, but they still struggle when applied to out-of-distribution examples. We address a challenging and underexplored version of this domain adaptation problem, where an algorithm is trained on several source domains, then examples from unseen domains that are unknown at training time. Particularly, no examples, labeled or unlabeled, any other knowledge about the target available present PADA: An example-based autoregressive Prompt learning for on-the-fly Any-Domain Adaptation, based T5 language model. Given test example, PADA first generates unique prompt it then, conditioned prompt, labels example with respect NLP prediction task. generate token sequence unrestricted length, consisting Domain Related Features (DRFs) characterize each domains. Intuitively, generated signature maps semantic space spanned by In experiments 3 tasks (text classification tagging), total 14 multi-source scenarios, substantially outperforms strong baselines.1