Daily Arxiv

This page organizes papers related to artificial intelligence published around the world.
This page is summarized using Google Gemini and is operated on a non-profit basis.
The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.

Mitigating Domain Shift in Federated Learning via Intra- and Inter-Domain Prototypes

Created by
  • Haebom

Author

Huy Q. Le, Ye Lin Tun, Yu Qiao, Minh NH Nguyen, Keon Oh Kim, Eui-Nam Huh, Choong Seon Hong

Outline

To address the heterogeneous domain problem where each client has a different feature distribution in a Federated Learning (FL) environment, this paper proposes I$^2$PFL, a novel Federated Prototype Learning method that utilizes both intra-domain and inter-domain prototypes. I$^2$PFL captures diversity within the local domain and improves generalization through feature alignment using MixUp-based augmented prototypes, and introduces a reweighting mechanism for inter-domain prototypes to provide cross-domain knowledge and reduce domain bias. Experiments on the Digits, Office-10, and PACS datasets demonstrate superior performance compared to existing methods.

Takeaways, Limitations

•
Takeaways:
◦
We present a novel approach that considers both intra-domain and inter-domain prototypes to solve heterogeneous domain problems in federated learning.
◦
Improved generalization performance through feature alignment and reweighting mechanisms using MixUp-based augmented prototypes.
◦
The superiority of the proposed method is demonstrated through experiments on various datasets.
•
Limitations:
◦
Only performance evaluations were conducted on specific datasets, and further research is needed to determine applicability in various real-world environments.
◦
Lack of analysis of computational complexity and communication costs.
◦
A more in-depth analysis of hyperparameter optimization of I$^2$PFL is needed.
👍
Made with Slashpage