ruixiaoLee / lut_seal_simulation

lut_seal_simulation

Our program is only tested on 64-bit platforms.
This project is based on the paper:
R. Li, H. Yamana: "Privacy-Preserving Function Evaluation using Lookup Tables with Word-Wise FHE," IEICE Trans. on Fundamentals of Electronics, Communications and Computer Sciences, vol.E107-A, No.8, pp.1-15,2024. doi:10.1587/transfun.2023EAP1114 (in printing)

Prerequisites

Microsoft SEAL version 4.0.0, OpenFHE 1.0.3, CMake, and OpenMP are needed.
Microsoft/SEAL
OpenFHE

Running

Revise the CMakeList.txt to compile the code you need.
When you make LUT use the source code makeTable.cpp. Please only available the part of the code you need.

cmake .
make

If you don't have directory ~/Table/two and ~/Table/three, please make them before you make LUTs for two and three-input functions evaluation.
Before running LUT processing, you have to generate keys and LUTs.

bin/keyGeneration
bin/makeTable

Note that you need to change the number of data points and threads in demo.hpp after you make LUTs, and compile the codes again.

Single-Input Function Evaluation with LUT

For one-input functions, you need to revise the LUT path in searchInput.cpp and extractOutput.cpp.

bin/makeInput
bin/searchInput
bin/makeQuery
bin/extractOutput
bin/checkResult

Multi-Input Function Evaluation with LUT

For two-input functions, you need to revise the LUT path in searchInputTwo.cpp and extractOutputTwo.cpp.

bin/makeInputTwo
bin/searchInputTwo
bin/makeQueryTwo
bin/extractOutputTwo
bin/checkResult

For three-input functions, you need to revise the LUT path in searchInputThree.cpp and extractOutputThree.cpp.

bin/makeInputThree
bin/searchInputThree
bin/makeQueryThree
bin/extractOutputThree
bin/checkResult

Comparison with Polynomial Approximation

To check the average abs/per error and runtime of the polynomial approximation method, run one of the codes you need.

bin/ckks_2(4/6/8)_relu(swish)

Comparison with Naive Bit-Wise LUT

Check the directory naivebitwiseLUT.

Construction of codes

src --- demo.hpp
      | keyGeneration.cpp
      | makeTable.cpp
      | checkResult.cpp(Step6)

      (For single-input:)
      | makeInput.cpp(Step1)
      | searchInput.cpp(Step2)
      | makeQuery.cpp(Step3)
      | extractOutput.cpp(Step4-5)

      (For two-input:)
      | makeInputTwo.cpp
      | searchInputTwo.cpp
      | makeQueryTwo.cpp
      | extractOutputTwo.cpp

      (For three-input:)
      | makeInputThree.cpp
      | searchInputThree.cpp
      | makeQueryThree.cpp
      | extractOutputThree.cpp

      (For polynomial approximation:)
      | ckks_2_relu.cpp
      | ckks_4_relu.cpp
      | ckks_6_relu.cpp
      | ckks_8_relu.cpp
      | ckks_2_swish.cpp
      | ckks_4_swish.cpp
      | ckks_6_swish.cpp
      | ckks_8_swish.cpp

      (For bit-wise FHE)
      | naivebitwiseLUT --- bitwiselut.cpp
                         |_ CMakeList.txt

      - randomInput --- test_normal.txt (test dataset)
                     |_ train_normal.txt (training dataset)
      - Table (store LUTs)
      - Result (store results)
      - compare_lut --- comError.py (compute the accuracy of LUT)
                     |_ genPoly.py (generate polynomial approximation)
                     |_ lut_ptxt_eds.py (generate plaintext LUT by eds)
                     |_ lut_ptxt_idas.py (generate plaintext LUT by ideas)