Federated optimization allows multiple institutions to jointly train a shared model on distributed data without centralizing the data. The new preprint focuses on the multimodal case, where the data includes medical images, genomic profiles, clinical records, and physiological signals. This heterogeneity introduces additional security and coordination challenges.
The paper proposes a hierarchical DNA-based key management framework to protect the federated optimization process. DNA-based cryptography uses biological sequences as a basis for encryption, and the hierarchical structure is designed to manage keys across different levels of the system. The abstract does not provide implementation details or experimental results.
Because this is a single source and the abstract is truncated, the claims here are limited to what the preprint states. No independent comparison is available.