← AI Techniques Radar
Inference efficiency

Latent Diffusion-based Obfuscation of Split Representations for Reconstruction-Resilient Medical Segmentation

Integrate a latent denoising diffusion model into the split point of a split federated learning pipeline so that the intermediate representations exchanged between client and server are diffused/denoised to reduce susceptibility to reconstruction attacks, while keeping the downstream segmentation task effective. The approach is explored via architectural variants that change how conditioning and feature fusion are applied around the diffusion module.

Discovered 2026-11-10 · Sources checked 2026-11-10

This summary was generated with AI assistance.

Why it matters

Split federated learning can leak sensitive information because transmitted intermediate features (and related signals) may be reconstructible. Obfuscating those representations at the split interface aims to mitigate reconstruction risk without abandoning collaborative training.

What changed

Instead of directly transmitting the raw intermediate features between client and server in SplitFed, the method inserts a latent diffusion denoising mechanism at the split boundary to transform/obfuscate the exchanged representation, with multiple variants testing different conditioning and fusion configurations.

Where it helps

  • Medical image segmentation with split federated learning
  • Privacy hardening against reconstruction attacks on exchanged intermediate features
  • Collaborative vision tasks where intermediate representations are sensitive

Evidence & reproduction

Reproduction not evaluated. A linked paper or repository does not establish independent reproduction.

Limitations & open questions

The described technique is presented in the context of medical image segmentation and is evaluated under specific reconstruction attack evaluations; the effectiveness and computational impact may vary with other data modalities, architectures, and threat models.

Sources