Federated learning for unpaired multimodal data through a homogeneous transformer model

Anders Eklund

Training of multimodal foundation models is currently restricted to centralized data centers containing massive, aligned datasets (e.g., image-text pairs). However, in realistic fe…