The IRMA Community
Newsletters
Research IRM
Click a keyword to search titles using our InfoSci-OnDemand powered search:
|
Privacy-Preserving Deep Learning for Healthcare Using Homomorphic Encryption
Abstract
Growing interest in cloud-based AI for clinical decision support raises critical patient privacy concerns, particularly under India's DPDPA. Homomorphic encryption (HE) offers a robust solution, enabling cloud servers to evaluate deep learning models on encrypted data and return encrypted results without accessing the plaintext. However, implementing HE requires navigating a strict privacy-efficiency-accuracy trilemma in deep learning architectures. This chapter provides a practical guide to designing neural network architectures that operate effectively under these cryptographic constraints. Per the authors, design choices such as polynomial activation function approximations, shallow-wide architectures, and quantization-aware training are discussed in detail. Through three healthcare case studies, the authors demonstrate that encrypted models can achieve accuracies within 1-3% of their plaintext counterparts. Finally, this chapter examines the integration of HE with federated learning and generative AI, concluding with a comprehensive implementation roadmap for healthcare teams.
Related Content
|
.
© 2027.
48 pages.
|
|
.
© 2027.
36 pages.
|
|
.
© 2027.
26 pages.
|
|
.
© 2027.
22 pages.
|
|
.
© 2027.
36 pages.
|
|
.
© 2027.
30 pages.
|
|
.
© 2027.
30 pages.
|
|
|