eduzhai > Applied Sciences > Engineering >

Domain Agnostic Internal Distributions for Unsupervised Model Adaptation

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 22 pages

Abstract: We develop an algorithm for sequential adaptation of a classifier that istrained for a source domain to generalize in a unannotated target domain. Weconsider that the model has been trained on the source domain annotated dataand then it needs to be adapted using the target domain unannotated data whenthe source domain data is not accessible. We align the distributions of thesource and the target domains in a discriminative embedding space via anintermediate internal distribution. This distribution is estimated using thesource data representations in the embedding space. We provide theoreticalanalysis and conduct extensive experiments on several benchmarks to demonstratethe proposed method is effective.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×