Federated Learning
How can we ensure the performance of all clients in FL? Is a server necessary in FL? How can multimodal fusion be more effective in FL?
Federated learning (FL) is an innovative collaborative approach that enables numerous clients to collectively enhance knowledge while prioritizing privacy, efficiency, and minimal communication overhead. Our endeavor centers on delivering personalized FL solutions, emphasizing a human-centric design. Beyond the traditional client-to-server framework, we are pioneering advancements in client-to-client, also known as decentralized FL. This encompasses the development of sophisticated communication protocols, network topologies, paradigms, and variants. Additionally, our development of a multimodal fusion FL approach significantly reduces communication costs and improves learning efficiency through selective modalities and client optimization. Our ultimate goal is to establish a fair, incentivized, and interpretable FL system, applicable in diverse areas including vehicular networks, academic research institutions, and mobile service platforms.
