Active Domain Adaptation via Clustering Uncertainty-weighted Embeddings

2021

Conference Paper

ps

Generalizing deep neural networks to new target domains is critical to their real-world utility. In practice, it may be feasible to get some target data labeled, but to be cost-effective it is desirable to select a subset that is maximally-informative via active learning (AL). In this work, we study the problem of AL under a domain shift. We empirically demonstrate how existing AL approaches based solely on model uncertainty or representative sampling are suboptimal for active domain adaptation. Our algorithm, Active Domain Adaptation via CLustering Uncertainty-weighted Embeddings (ADA-CLUE), i) identifies diverse datapoints for labeling that are both uncertain under the model and representative of unlabeled target data, and ii) leverages the available source and target data for adaptation by optimizing a semi-supervised adversarial entropy loss that is complimentary to our active sampling objective. On standard image classification benchmarks for domain adaptation, ADA-CLUE consistently performs as well or better than competing active adaptation, active learning, and domain adaptation methods across shift severities, model initializations, and labeling budgets.

@inproceedings{Prabhu:ICCV:2021,
  title = {Active Domain Adaptation via Clustering Uncertainty-weighted Embeddings},
  author = {Prabhu, Viraj and Chandrasekaran, Arjun and Saenko, Kate and Hoffman, Judy},
  booktitle = {Proc. International Conference on Computer Vision (ICCV)},
  pages = {8485--8494},
  publisher = {IEEE},
  address = {Piscataway, NJ},
  month = oct,
  year = {2021},
  doi = {10.1109/ICCV48922.2021.00839},
  month_numeric = {10}
}