PPLR

C++ implementation of PPLR

Summary

PPLR is a library for computing Logistic Regression in privacy-preserving manner with Homomorphic Encryption. It contains two applications CLR and SLR.

• CLR : Secure version (Primary submission)
• SLR : Fast version (Secondary submission)

List of Supported CPUs and Operating Systems

• Intel 64-bit CPUs
• 64-bit Linux (tested on ubuntu 14.04 LTS and ubuntu 16.04 LTS)

List of Supported Compilers and Prerequisite Tools

• gcc 4.8.4 + OpenSSL 1.0.1f + GMP 5.1.3

Installation Requirements

Create a working directory (e.g., work) and clone the following repositories.

	cd ~
	mkdir work
	cd work
	git clone git://github.com/cBioLab/PPLR.git
	git clone git://github.com/iskana/PBWT-sec.git
	git clone git://github.com/herumi/xbyak.git
	git clone git://github.com/herumi/mcl.git
	git clone git://github.com/herumi/cybozulib.git
	git clone git://github.com/herumi/cybozulib_ext.git

• Xbyak is a prerequisite for optimizing the operations in the finite field on Intel CPUs.
• OpenSSL and libgmp-dev are available via apt-get (or other similar commands).

Installation

   cd mcl
   git checkout 61473db5730731327059fa42a445853d853c6a21
   cd ..
   cd cybozulib
   git checkout 94b914a1d93b3ecf7f7aa096006d8a72b39c7496
   cd ..
   cd cybozulib_ext
   git checkout fe3bfe014e72c779433e8b02bef9066e2bfec9db
   cd ..
   cd xbyak
   git checkout c5da3778e7f84013fe8c26fcf18a67881bd1e825
   cd ../PPLR/comm
   mkdir server
   mkdir client
   cd ../src
   make

Prerequisite Files and Libraries for Running Your Application

* OpenSSL
* GMP (libgmp-dev)

Quick start

Running CLR sample

cd ../bin
./clr.sh

Running SLR sample

cd ../bin
./slr.sh

Running CLR

precomputation (key generation instead of analyst, key distribution, data encryption instead of data providers) 
./encdata [training data file]

@server
./clrserver -p [port number] -c [the number of threads]

@client
./clrclient -h [ip address of server] -p [port number] -i [the number of learning iterations] -l [learning rate] -r [coefficient of L2 norm] -a [accuracy]

• This demonstration shows the execution procedure of CLR after the server receives the encrypted data from the data providers. For convenience, key generation, key distribution and data encryption is executed beforehand.
• Following files are created by precomputation.
  • Encrypted data : PPLR/data/encdata.dat.
  • PublicKey data : PPLR/data/pubsample.dat.
  • SecretKey data : PPLR/data/prvsample.dat.
• Before running the server, you put Encrypted data and PublicKey data in PPLR/data/ of the server.
• Before running the client, you put SecretKey data and PublicKey data in PPLR/data/ of the client.
• If accuracy is s, the logistic regression model parameters are calculated with s digits after the decimal point.

Running SLR

precomputation (key generation instead of analyst, key distribution, data encryption instead of data providers) 
./slrdatap -f [training data file]

@server
./slrserver -p [port number]

@client
./slrclient -p [port number] -h [ip address of the server]
python slr.py [learning rate] [coefficient of L2 norm]

• This demonstration shows the execution procedure of SLR after the server receives the encrypted data from the data providers. For convenience, key generation, key distribution and data encryption is executed beforehand.
• Following files are created by precomputation.
  • Encrypted data : PPLR/data/encdata.dat.
  • PublicKey data : PPLR/data/slrpub.dat.
  • SecretKey data : PPLR/data/slrprv.dat.
• Before running the server, you put Encrypted data and PublicKey data in PPLR/data/ of the server.
• Before running the client, you put SecretKey data and PublicKey data in PPLR/data/ of the client.

Converting from sample-data to training data file and test data file

	cd PPLR/data
	python convert.py [sample data e.g."10-5.csv"] [training data file] [test data file]

Training data format

   	1st line   : The number of data providers
   	2nd line   : Dimension of data 
   	3rd line ~ : Data
	(i-th line : Data of (i-2)-th data providers [y x1 x2 ... xD])

Copyright Notice

Copyright (C) 2017, Masanobu Jimbo All rights reserved.

License

PPLR (files in this repository) is distributed under the [BSD 3-Clause License] (http://opensource.org/licenses/BSD-3-Clause "The BSD 3-Clause License").

Licenses of External Libraries

Licenses of external libraries are listed as follows.

• PBWT-sec: BSD-3-Clause
• cybozulib: BSD-3-Clause
• mcl: BSD-3-Clause
• Xbyak: BSD-3-Clause
• MPIR: LGPL2
• OpenSSL: Apache License Version 1.0 + SSLeay License

Software including any of those libraries is allowed to be used for commercial purposes without releasing its source code as long as the regarding copyright statements described as follows are included in the software.

• This product includes software that was developed by an OpenSSL project in order to use OpenSSL toolkit.
• This product includes PBWT-sec.
• This product includes mcl, cybozulib, and Xbyak.
• This product includes MPIR.

Contact Information

• Masanobu Jimbo (jimwase@asagi.waseda.jp)

History

2017/August/16; initial version