Peregrine: One-Shot Fine-Tuning for FHE Inference of General Deep CNNs
This work addresses the problem of efficient and secure inference for deep learning models in privacy-sensitive applications, representing a domain-specific advancement in FHE for CNNs.
The paper tackles the challenges of adapting deep CNNs for fully homomorphic encryption (FHE) inference by proposing a single-stage fine-tuning strategy and a generalized interleaved packing scheme, achieving competitive accuracy on datasets like CIFAR-10 and ImageNet and enabling the first FHE-based inference for YOLO object detection.
We address two fundamental challenges in adapting general deep CNNs for FHE-based inference: approximating non-linear activations such as ReLU with low-degree polynomials while minimizing accuracy degradation, and overcoming the ciphertext capacity barrier that constrains high-resolution image processing on FHE inference. Our contributions are twofold: (1) a single-stage fine-tuning (SFT) strategy that directly converts pre-trained CNNs into FHE-friendly forms using low-degree polynomials, achieving competitive accuracy with minimal training overhead; and (2) a generalized interleaved packing (GIP) scheme that is compatible with feature maps of virtually arbitrary spatial resolutions, accompanied by a suite of carefully designed homomorphic operators that preserve the GIP-form encryption throughout computation. These advances enable efficient, end-to-end FHE inference across diverse CNN architectures. Experiments on CIFAR-10, ImageNet, and MS COCO demonstrate that the FHE-friendly CNNs obtained via our SFT strategy achieve accuracy comparable to baselines using ReLU or SiLU activations. Moreover, this work presents the first demonstration of FHE-based inference for YOLO architectures in object detection leveraging low-degree polynomial activations.