Domain Adaptation
Closing the gap when a model trained on one data distribution gets deployed on a shifted one, without collecting new labels for the new distribution.
Think of It Like This
Like an experienced English soccer referee learning to officiate American football games by transferring their foul-calling instincts to new rules.
Covariate shift means the input distribution changes between the source and target domains while the true input-output relationship stays fixed — unlike concept shift, where the task's actual definition changes. Domain-adversarial training pushes internal features toward a form a domain classifier can't distinguish, so behavior learned on the source domain transfers to the target one.
It's typically framed as unsupervised: the target domain has no labels at all to fine-tune on.