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.

MMiC: Mitigating Modality Incompleteness in Clustered Federated Learning

Created by
  • Haebom

Author

Lishan Yang, Wei Emma Zhang, Quan Z. Sheng, Lina Yao, Weitong Chen, Ali Shakeri

Outline

This paper proposes a novel framework, MMiC, to address the problem of missing modality in multimodal federated learning (MFL), a distributed learning method that utilizes data with diverse modalities. MMiC mitigates the impact of missing modality by replacing partial parameters of client models within a cluster, optimizes client selection using the Banzhaf Power Index, and dynamically controls global aggregation using Markovitz Portfolio Optimization. Experimental results demonstrate that MMiC outperforms existing federated learning architectures in both global and personalized performance on multimodal datasets with missing modality.

Takeaways, Limitations

•
Takeaways:
◦
We present a novel framework, MMiC, that effectively addresses the missing modality problem in multimodal federated learning.
◦
Client selection and global aggregation optimization using Banzhaf Power Index and Markovitz Portfolio Optimization.
◦
We demonstrate improved performance over existing methods on multimodal datasets with missing modalities.
◦
Ensure reproducibility and usability through open code.
•
Limitations:
◦
The performance of the proposed method may depend on specific datasets and settings.
◦
Generalization performance evaluation is needed for various modality types and missing patterns.
◦
Further research is needed on scalability and efficiency in real-world application environments.
👍
Made with Slashpage