In this repository, you can find the implementation of self-supervised Fisher merging, a new model aggregation technique (https://openreview.net/forum?id=sPtEDSVD4K) that optimizes parameters without re-visiting data. This is the result from a collaboration between Barcelona Supercomputing Center (BSC) and Technical University of Denmark (DTU).
Please, if you use this code, include the following citation:
@inproceedings{
recasens2023beyond,
title={Beyond Parameter Averaging in Model Aggregation},
author={Pol G. Recasens and Jordi Torres and Josep Berral and S{\o}ren Hauberg and Pablo Moreno-Mu{\~n}oz},
booktitle={International Workshop on Federated Learning in the Age of Foundation Models in Conjunction with NeurIPS 2023},
year={2023},
url={https://openreview.net/forum?id=sPtEDSVD4K}
}