FEDALV

Federated Active Learning for Domain Generalisation

This is the official code release for FEDALV.

Paper: https://arxiv.org/pdf/2312.02247.pdf

Datasets used:

FEDALV Pipeline

Active Learning is integrated at the server level in our proposed Federated Domain Generalisation framework. Initially, a subset of source data is labelled, and over time, it is refined to retain only the most informative samples for target generalisation. Communication links with the global model operate during the Federated Learning training, and Active Learning queries are propagated once the process is complete. After just a couple of cycles, the target domain (light blue) aligns with the sources in class clusters.

Deployment

We recommend to create a python3 environment and install the requirements.txt file

pip install -r requirements.txt

Run

Contact Info

Razvan Caramalau: r.caramalau@ucl.ac.uk

Acknowledgement

This work was supported in whole, or in part, by the Wellcome/EPSRC Centre for Interventional and Surgical Sciences (WEISS) [203145/Z/16/Z], the Department of Science, Innovation and Technology (DSIT) and the Royal Academy of Engineering under the Chair in Emerging Technologies programme; Horizon 2020 FET Open (EndoMapper) (12/2019 – 11/2024) [863146]. Not for commercial use. Also, thankful to the following repos:

Citation

Caramalau, R., Bhattarai, B., and Stoyanov, D., “Federated Active Learning for Target Domain Generalisation”, arXiv e-prints, 2023. https://doi.org/10.48550/arXiv.2312.02247

BibTeX

@misc{caramalau2023federated,
      title={Federated Active Learning for Target Domain Generalisation}, 
      author={Razvan Caramalau and Binod Bhattarai and Danail Stoyanov},
      year={2023},
      eprint={2312.02247},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

razvancaramalau / fedalv