Efficient Secure Federated Learning via Information-Theoretically Secure Key Distribution: A Medical Imaging Case Study
Secure federated learning cuts information-theoretic key consumption about 35-fold for chest X-ray training.
The paper addresses secure aggregation for federated learning when additive masking must use physics-based, information-theoretically secure keys with limited generation rate and storage. Frozen backbones, knowledge distillation, and quantization shrink model-update payloads and therefore key-material consumption. On a real key-distribution testbed for chest X-ray classification, key usage fell by about 35 times while predictive accuracy was maintained, avoiding buffer depletion and key expiration.
- Additive masking for federated updates needs scarce physical keys.
- Frozen backbones, distillation, and quantization shrink payloads.
- Real testbed used chest X-ray classification and physical keys.
- Key use fell about 35 times with accuracy maintained.
From the original article
1 min read at arxiv.orgFederated Learning (FL) enables collaborative training of models across institutions without centralizing sensitive data, making it well-suited for privacy-concerned applications, such as medical imaging. To protect FL model updates during secure aggregation, additive masking is commonly employed. However, its underlying classical key establishment is only computationally secure.
By Ivan Donà · 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.