[공지사항]을 빙자한 안부와 근황 
Show more

Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

FeDa4Fair: Client-Level Federated Datasets for Fairness Evaluation

Created by
  • Haebom

Author

Xenia Heilmann, Luca Corbucci, Mattia Cerrato, Anna Monreale

Outline

In this paper, we present FeDa4Fair, a library for benchmarking fair FL methods under heterogeneous client data distributions, and four datasets and benchmarks with heterogeneous biases to address the fairness issue in Federated Learning (FL). Unlike previous studies that focus on a single binary sensitive attribute, FeDa4Fair supports more robust and reproducible fairness studies by considering diverse and potentially conflicting fairness demands from clients. FeDa4Fair generates tabular datasets for evaluating fair FL methods under various client biases and provides functions for evaluating fairness results.

Takeaways, Limitations

•
Takeaways:
◦
A novel benchmarking framework for addressing fairness issues in federated learning.
◦
Provides diverse and realistic datasets that take into account heterogeneous client biases.
◦
Supporting reproducible research by providing convenient functions for fairness assessment.
◦
Presenting research possibilities that take into account the conflicting fairness requirements of various clients.
•
Limitations:
◦
Further validation of the generalizability of the presented dataset is needed.
◦
The currently provided dataset is limited to tabular data. It needs to be extended to other data types (images, text, etc.).
◦
Lack of comprehensive support for various fairness concepts and metrics.
◦
Lack of consideration of complex sensitive properties beyond single sensitive properties.
👍
Made with Slashpage