<ul><li>Multimodal Federated Learning (MFL) combines leveraging multiple modalities to enhance downstream inference performance and enabling distributed training for efficiency and privacy preservation.</li><li>A comprehensive taxonomy for organizing MFL within different Federated Learning (FL) paradigms is currently lacking, despite the growing interest in MFL.</li><li>Challenges in MFL, such as modality heterogeneity and communication inefficiency, differ significantly from those in traditional unimodal scenarios.</li><li>This paper systematically examines MFL in the context of horizontal FL (HFL), vertical FL (VFL), and hybrid FL paradigms, discussing challenges and providing insights for future research.</li></ul>

Multimodal Federated Learning: A Survey through the Lens of Different FL Paradigms

Discover more