FedAMB: Adaptive Modality Balancing for Dominance-Robust Multimodal Federated Distillation
Experiments show that FedAMB improves multimodal accuracy and missing-modality robustness and prevents the propagation of fusion-biased teachers while directly improving unimodal representations.
Seung-Hwa Han, Juyeob Lee, Sang-Min Lee et al.
· 0 citations