Paper proposes DNA-domain encryption for multimodal federated learning
Original title: Secure Multimodal Federated Optimization via Hierarchical DNA-Based Key Management
Clients derive per-modality keys locally, and the coordinator sees only securely aggregated updates.
The paper outlines federated optimization over medical images, genomic profiles, clinical records, and physiological signals without centralizing raw data. A key authority provisions one client key with HPKE; clients derive modality keys with HKDF, map data into a DNA-domain form, and encrypt with AES-GCM. Only required modalities are decrypted locally, and model updates use secure aggregation.
- One HPKE client key yields modality-specific symmetric keys via HKDF.
- Modalities undergo a DNA-inspired transform before AES-GCM encryption.
- The coordinator observes aggregated updates, not each client's contribution.
From the original article
1 min read at arxiv.orgFederated optimization over heterogeneous multimodal data enables distributed institutions to collaboratively optimize a shared model using medical images, genomic profiles, clinical records, and physiological signals without centralizing raw data.
By Yuping Yan · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.