Previous
Privacy-Preserving Federated Medical Imaging with Encrypted Representations
Conventional federated learning avoids sharing raw medical images, but transmitted gradients or model updates can still leak sensitive patient information through reconstruction and inversion attacks. Encrypting entire gradients can also impose substantial communication and computation overhead.
Project Summary
This work explored a privacy-preserving federated learning design using compact Vision Transformer representations rather than full model gradients. Homomorphic encryption was applied to protected feature representations so that collaborative medical-image learning could reduce reconstruction exposure while lowering communication overhead.