Skip to content

Towards Case-Optimized Hybrid Homomorphic Encryption

Year: 2022

type: technique collection: schemes status: auto

Links: ePrint 2022/180

Summary

Auto-extracted summary (unreviewed)

Introduces the Group Filter Permutator paradigm, generalizing Improved Filter Permutators, and the Elisabeth family of HE-friendly stream ciphers, instantiated as Elisabeth-4 with 4-bit components tailored to TFHE's programmable bootstrapping. A Rust implementation over Concrete transciphers at about 91 ms per ciphertext on a 64-core Threadripper, comparable to state-of-the-art HHE, and the authors show that data sent through Elisabeth-4 can feed homomorphic neural-network inference on Fashion-MNIST with no accuracy loss relative to fresh FHE ciphertexts and large bandwidth savings.

Abstract

Hybrid Homomorphic Encryption (HHE) reduces the amount of computation client-side and bandwidth usage in a Fully Homomorphic Encryption (FHE) framework. HHE requires the usage of specific symmetric schemes that can be evaluated homomorphically efficiently. In this paper, we introduce the paradigm of Group Filter Permutator (GFP) as a generalization of the Improved Filter Permutator paradigm introduced by M´eaux et al. From this paradigm, we specify Elisabeth , a family of stream cipher and give an instance: Elisabeth-4 . After asserting the security of this scheme, we provide a Rust implementation of it and ensure its performance is comparable to state-of-the-art HHE. The true strength of Elisabeth lies in the available operations server-side: while the best HHE applications were limited to a few multiplications server-side, we used data sent through Elisabeth-4 to homomorphically evaluate a neural network inference. Finally, we discuss the improvement and loss between the HHE and the FHE framework and give ideas to build more efficient schemes from the Elisabeth family.
Source: pdf

Extracted details

Field Value Source Confidence
contribution Group Filter Permutator (GFP) paradigm generalizing the Improved Filter Permutator llm 0.85
contribution Elisabeth stream cipher family and the Elisabeth-4 instance over Z16 llm 0.85
contribution Security assessment of Elisabeth-4 at 128-bit security llm 0.85
contribution Rust implementation over Concrete, plus re-implementation of FiLIP for fair comparison llm 0.85
contribution First HHE use case with homomorphic NN inference (Fashion-MNIST) on transciphered data llm 0.85
schemes TFHE llm 0.85
schemes hybrid homomorphic encryption llm 0.85
schemes Elisabeth-4 stream cipher llm 0.85
schemes FiLIP (compared) llm 0.85
schemes PASTA/HERA/LowMC/Kreyvium/RASTA/DASTA/MASTA (compared) llm 0.85
libraries Concrete llm 0.85
libraries Rust llm 0.85
models neural network classifier (Fashion-MNIST, 3-bit quantized inputs) llm 0.85
datasets Fashion-MNIST llm 0.85
tasks transciphering llm 0.85
tasks image classification llm 0.85
hardware AMD Ryzen Threadripper 3990X 64-core CPU llm 0.85
security level 128-bit llm 0.85
threat model client-server llm 0.85
interaction non-interactive FHE llm 0.85
code url https://github.com/princess-elisabeth/sboxes (URL found in the full text) regex 0.50
code url https://www.github.com/princess-elisabeth/Elisabeth (URL found in the full text) regex 0.50
code url https://github.com/princess-elisabeth/FiLIP (URL found in the full text) regex 0.50
code url https://github.com/princess-elisabeth/Elisabeth llm 0.85
limitations Use case is simple (3-bit quantized Fashion-MNIST); finer-grained data may need larger NLUTs or more complex models llm 0.85
limitations Stand-alone homomorphic evaluation of Elisabeth-4 only matches state-of-the-art competitors llm 0.85
limitations Cross-cipher benchmark comparisons must be read with care because ciphers target different FHE schemes llm 0.85
llm relevance low llm 0.85
key result transciphering time per ciphertext: 91.143 ms (two KS), 103.810 ms (single KS) (Elisabeth-4 over Concrete, 64-core Threadripper, multithreaded) llm 0.85
key result transciphering time per bit: 22.786 ms (Elisabeth-4 two KS, multithreaded) llm 0.85
key result transciphering time per ciphertext (single thread): 1485.0 ms (Elisabeth-4 two KS) llm 0.85
key result inference accuracy: unchanged vs direct FHE on fresh ciphertexts (Fashion-MNIST homomorphic classification through Elisabeth-4) llm 0.85

Keyword mentions

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

  • scheme: Hybrid HE / transciphering 52 [p.1], TFHE 38 [p.2], CKKS 3 [p.14], GSW 3 [p.5], Paillier 3 [p.28], MPC 2 [p.28]
  • library: Concrete (Zama) 16 [p.4], HElib 7 [p.14], Zama 1 [p.1]
  • model: CNN 6 [p.24]
  • dataset: MNIST 6 [p.2], Fashion-MNIST 1 [p.2]
  • technique: Programmable bootstrapping 34 [p.2], Bootstrapping 16 [p.2], Lookup table 14 [p.2], Quantization 5 [p.2], Rescaling 3 [p.5], Sigmoid 3 [p.24], Convolution 2 [p.24]
  • assumption: LWE 30 [p.4], RLWE 4 [p.4], Lattice-based 3 [p.30], 128-bit security 2 [p.11]
  • task: Medical / healthcare 2 [p.24]
  • topic: Homomorphic encryption 32 [p.1], Benchmark 10 [p.2], Open source 5 [p.11]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/schemes/s42-3_eprint-2022-180.pdf schemes 35 498.0 KB a9ad96afdf03

Database key: towards-case-optimized-hybrid-homomorphic-encryption · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set towards-case-optimized-hybrid-homomorphic-encryption COLUMN VALUE