AutoPoly-HE: Autonomous Design of Polynomial Activations for Privacy-Preserving Encrypted Inference
AutoPoly-HE automatically designs polynomial activations for encrypted inference across ResNet and Vision Transformer models.
AutoPoly-HE automatically selects polynomial activation functions for homomorphic-encryption inference, replacing hand-designed approximations that ignore dataset and architecture differences. The authors evaluate ResNet-18, ResNet-34, ResNet-50, ViT-S/16, ViT-B/16, and ViT-L/16 across multiple datasets. Transformer models keep higher accuracy under polynomial approximation, while CNN accuracy degrades more as tasks grow harder. A CLS-token inference pipeline removes depth-dependent operations, leaving encryption and decryption as the main latency cost.
- AutoPoly-HE selects polynomial activations for each dataset and model.
- Tested ResNet-18/34/50 and ViT-S/16, ViT-B/16, and ViT-L/16.
- Transformers retained higher accuracy under approximation than CNNs.
- CLS-token inference shifts cost mainly to encryption and decryption.
- Tags
- homomorphic-encryptionprivacy-preserving-mlpolynomial-activationresnetvision-transformerencrypted-inferenceautoply-he
- Products
- ResNetVision Transformer
From the original article
1 min read at arxiv.orgPrivacy-preserving machine learning has become critical for deploying artificial intelligence in sensitive domains such as healthcare, finance, and infrastructure monitoring. Homomorphic encryption (HE) enables secure inference on encrypted data. Despite its advantages, HE is not compatible with nonlinear activation functions found in machine learning models.
By Zeinab Elkhatib · 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.