RASO
A server-side optimization method for multimodal federated learning that improves model performance under client-side modality missing scenarios through role-aware contributor selection and component-specific momentum strategies.
Timeline (last 90 days)
RASO (Role-Aware Server Optimization) is a multimodal federated learning method for missing-modality scenarios that improves model performance by coordinating contributor selection with role-specific server updates.
RASO refines the server-side aggregation operation into two roles: the encoder role and the prediction block role, and formulates update strategies for each separately.
RASO's contributor selection selects contributors based on prediction dependency and participation in local optimization.
For the encoder role, RASO retains server momentum to stabilize the encoder's adaptation process, even when certain modalities are missing on some clients.
For the shared prediction block, RASO synchronizes it to the current aggregation result without using momentum, thereby avoiding the possible negative effects of momentum.
Experimental results show that by coordinating contributor selection with role-specific server updates, RASO can significantly improve model performance under missing-modality conditions.
All Facts (6)
RASO (Role-Aware Server Optimization) is a multimodal federated learning method for missing-modality scenarios that improves model performance by coordinating contributor selection with role-specific server updates.
50%UnverifiedRASO refines the server-side aggregation operation into two roles: the encoder role and the prediction block role, and formulates update strategies for each separately.
50%UnverifiedRASO's contributor selection selects contributors based on prediction dependency and participation in local optimization.
50%UnverifiedFor the encoder role, RASO retains server momentum to stabilize the encoder's adaptation process, even when certain modalities are missing on some clients.
50%UnverifiedFor the shared prediction block, RASO synchronizes it to the current aggregation result without using momentum, thereby avoiding the possible negative effects of momentum.
50%UnverifiedExperimental results show that by coordinating contributor selection with role-specific server updates, RASO can significantly improve model performance under missing-modality conditions.
50%