Researchers have developed a new method for replacing the ReLU activation function in neural networks with quadratic polynomials, specifically for use with fully homomorphic encryption (FHE). This approach aims to reduce the computational cost of FHE-only inference by using lower-degree polynomials while preserving classification accuracy on calibration datasets. The method formulates the replacement as a linear separation problem and extends to cases with misclassified samples using convex hull relaxations, achieving faster inference times compared to existing methods. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Enables more efficient inference for neural networks using fully homomorphic encryption, potentially reducing costs and increasing adoption.
RANK_REASON Academic paper detailing a novel technical approach. [lever_c_demoted from research: ic=1 ai=1.0]