FedMGP: Personalized Federated Learning with Multi-Group Text-Visual Prompts

FedMGP is a novel method in personalized federated learning designed to improve vision-language models by leveraging multiple groups of paired text and visual prompts provided to clients. This multi-group approach enables the capture of diverse semantic details across different prompt sets, enhancing the model's ability to understand nuanced information. A key component of FedMGP is the introduction of a diversity loss function, which encourages each prompt group to focus on distinct aspects of the data, thereby reducing redundancy and promoting richer feature representation. By integrating these elements, FedMGP aims to create more effective and personalized models within federated learning frameworks. This approach reflects ongoing advancements in combining text and visual modalities for improved machine learning performance. The method was detailed in a recent publication on arXiv in November 2025, highlighting its relevance to current AI research.