Optimized layerwise approximation for efficient private inference on fully homomorphic encryption

doi:10.1016/j.neucom.2026.134016

Optimized layerwise approximation for efficient private inference on fully homomorphic encryption

2026-05-19

0

PRE

AI

DOI：10.1016/j.neucom.2026.134016

Original

Original request for help

Save

Abstract

En 中文

• Fully homomorphic encryption strongly protects sensitive data. • Fully homomorphic encryption requires approximations for non-arithmetic operations. • Post-training approximation eliminates the need for retraining the pre-trained model. • Evaluating high-degree polynomials on ciphertexts incurs significant time latency. • Optimized layerwise approximation reduces time latency for private inference.

Keywords:

Fully homomorphic encryption

Private inference

Layerwise approximation

Time latency

Non-arithmetic operations

Optimized layerwise approximation for efficient private inference on fully homomorphic encryption

Abstract

Journal

Organization