Skip to content

Swift: Fast Secure Neural Network Inference With Fully Homomorphic Encryption

Authors: Yu Fu; Yu Tong; Yijing Ning; Tianshi Xu; Meng Li; Jingqiang Lin; Dengguo Feng
Year: 2025
Venue: IEEE Transactions on Information Forensics and Security

type: system collection: ppml status: auto

Links: DOI

Summary

Auto-extracted summary (unreviewed)

Swift is a secure two-party neural-network inference scheme that uses FHE with coefficient encoding for linear layers and combines FHE (SIMD encoding) with secret sharing to build efficient protocols for ReLU and max pooling, together with a conversion protocol between coefficient-encoded and SIMD-encoded ciphertexts. On MNIST inference in a WAN setting, Swift improves the execution time of ReLU, max pooling and end-to-end secure inference by 7.4x, 13.3x and 1.9x over Cheetah.

Abstract

With the widespread use of machine learning (ML), privacy concerns during neural network inference are attracting growing attention. Secure two-party neural network (2PC-NN) inference is the privacy-preserving inference method, which allows client to obtain the inference result without disclosing client’s input to the server. The server’s model parameters are also confidential to the client. However, current 2PC-NN inference schemes still have large overhead, especially for non-linear functions. In this paper, we present Swift, a fast secure 2PC- NN inference scheme based on fully homomorphic encryption (FHE) and secret sharing (SS). FHE protects the input and model parameters in linear functions, while SS is integrated to protect the non-linear functions. Concretely, Swift integrates FHE and SS to design secure and efficient non-linear protocols used for ReLU and max pooling. To further optimize performance, Swift employs FHE with computation-friendly coefficient encoding for fast execution of linear functions, and SIMD encoding for nonlinear functions. Swift constructs efficient encoding conversion protocol between the coefficient-encoded ciphertext and the SIMD-encoded ciphertext. Finally, Swift achieves secure neural network inference framework for MNIST dataset. Compared with Cheetah (USENIX 2022), the execution time of ReLU, max pooling, secure inference under a WAN setting improves 7.4×, 13.3×, 1.9×, respectively.
Source: pdf

Extracted details

Field Value Source Confidence
contribution FHE + secret-sharing protocols for non-linear ReLU and max pooling llm 0.70
contribution coefficient encoding for fast linear layers and SIMD encoding for non-linear layers llm 0.70
contribution encoding conversion protocol between coefficient-encoded and SIMD-encoded ciphertexts llm 0.70
contribution end-to-end secure 2PC neural network inference framework evaluated on MNIST llm 0.70
schemes FHE llm 0.70
schemes secret sharing llm 0.70
libraries SEAL llm 0.70
libraries EMP-toolkit llm 0.70
models neural network with ReLU and max pooling (architecture not specified in excerpt) llm 0.70
datasets MNIST llm 0.70
tasks private neural network inference llm 0.70
tasks image classification llm 0.70
security level 128-bit llm 0.70
threat model two-party client-server (2PC-NN); client input and server model parameters kept confidential llm 0.70
interaction interactive (HE+MPC) llm 0.70
code url https://github.com/Microsoft/SEAL (URL found in the full text) regex 0.50
code url https://github.com/Alibaba-Gemini-Lab/OpenCheetah (URL found in the full text) regex 0.50
llm relevance medium llm 0.70
key result ReLU speed-up over Cheetah: 7.4x (WAN) llm 0.70
key result max pooling speed-up over Cheetah: 13.3x (WAN) llm 0.70
key result end-to-end secure inference speed-up over Cheetah: 1.9x (MNIST, WAN) llm 0.70

Keyword mentions

Counts of vocabulary terms in the full text (first page of occurrence in brackets).

  • scheme: CKKS 18 [p.2], MPC 6 [p.1], TEE / confidential computing 5 [p.2], TFHE 2 [p.2], FHEW 2 [p.2], Scheme switching 2 [p.13], Secret sharing 2 [p.1]
  • library: Microsoft SEAL 5 [p.2], ABY / ABY3 3 [p.10], EzPC / CrypTFlow 3 [p.10], nGraph-HE 2 [p.2], TensorFlow 1 [p.13]
  • model: CNN 4 [p.9], ResNet 1 [p.6]
  • dataset: MNIST 4 [p.1]
  • hardware: AMD EPYC 1 [p.9]
  • technique: ReLU 67 [p.1], SIMD packing 39 [p.1], Two-party computation 19 [p.1], Convolution 17 [p.2], Bootstrapping 9 [p.6], Semi-honest model 6 [p.3], Rotations 5 [p.1], RNS 4 [p.2], Malicious model 4 [p.9], Model extraction / stealing 2 [p.9]
  • assumption: RLWE 8 [p.2], LWE 2 [p.2]
  • topic: Homomorphic encryption 60 [p.1], Private / secure inference 9 [p.1], Open source 4 [p.9], Benchmark 2 [p.9]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/ppml/inference/fu-2025_swift-fast-secure-neural-network-inference-with-fhe.pdf ppml/inference 14 2.8 MB bbea57c555f9 was Fu et al. - 2025 - Swift Fast Secure Neural Network Inference With Fully Homomorphic Encryption.pdf

Database key: swift-fast-secure-neural-network-inference-with-fully-homomorphic · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set swift-fast-secure-neural-network-inference-with-fully-homomorphic COLUMN VALUE