Heterogeneous network data-free fusion method and system based on federated distillation

The invention discloses a heterogeneous network data-free fusion method and system based on federated distillation, and belongs to the technical field of information processing. For information among different combat systems, a CGAN model is trained through a federated learning method and is used for optimizing local data, generating a training set with independent identical distribution characteristics, and improving the model training efficiency and precision; a transfer set for distillation is generated by utilizing the CGAN network instead of migrating a small amount of samples from source data, so that the confidentiality requirement of the data is met, and local model knowledge is migrated in a data-free manner; a federated distillation method is utilized to aggregate the local model, the isomorphism requirements of the local model and a global model in a traditional federated learning algorithm are weakened, an edge server with data can pertinently design the local model according to a local data struc