sls33 / SecureKVM

SecureKVM: A MPC-based framework for efficient privacy preserving learning of key-value models

SecureKVM

Introduction

SecureKVM is a branch of SecMML that focuses on enabling secure and efficient inference (or training) for key-value models. Currently, we only support key-value models like Markov Model and Decision Tree Model.

Running

Network Configuration

Default setting

IP:port

127.0.0.1:1234

In the default configuration, multiple processes run on a server, which is a LAN environment.

In order to simulate WAN environment, we can use the following command.

# simulate WAN
sudo tc qdisc add dev lo root netem delay 20ms rate 40Mbps
# cancle the simulation
sudo tc qdisc del dev lo root netem delay 20ms rate 40Mbps
# tc qdisc add dev <dev> root <qdisc> <qdisc-param>

How to configure

SecureKVM/util/SocketManager.cpp line127

void SocketManager::SMMLF::init() {
    init("127.0.0.1", 1234); // ip, port
}

Compile

Constant.h

#define M 3
#define OFFLINE_PHASE_ON true
#define TOTAL_SAMPLE_SIZE 10

• M: the n in n-party
• OFFLINE_PHASE_ON: whether running offline phase or not
• TOTAL_SAMPLE_SIZE: the number of tests

CMakeLists.txt

explanation of executable file names

• test_xxx: test benchmark of operation ( SecGetPer Log)
• xxx_func: test function of basic operation (Reshare GenPerm ShuffleData SecGetPer Log)
• DT_MPC: Secure Decision Tree Inference
• evalPW_mpc: Secure Markov Inference

Compile the project at the directory SecureKVM/

cmake .
make -j

Test Basic Operation

Constant.h

• PREDICTION_BATCH_BENCHMARK: the bacth size

Run

./executable_filename partynum

For example, to test protocol secgetper in three-party setting, you need open three terminals and execute the following commands in turn.

./test_secgetper 0
./test_secgetper 1
./test_secgetper 2

partynum should start from 0, end at M-1.

Please make sure the M in constant.h is right.

If you want to test the correctness of the protocol, please see the file SecureKVM/test_op/test_genperm.cpp as example.

Test Inference

Constant.h

set BENCHMARK to false;

1 Secure Decision Tree Inference

Constant.h

if PERMUTE_USING_BASIC is true, use Secget_Basic protocol, else using Secget_Fold protocol.

FEATURE_DIM: the feature size of dataset.

DEPTH: the depth of decision tree.

PREDICTION_BATCH: the batch size of inference

#define FEATURE_DIM 9
#define PREDICTION_BATCH 10
#define DEPTH 8
#define NODE_NUM ((1 << DEPTH) - 1)
#define PERMUTE_USING_BASIC true
#if (PERMUTE_USING_BASIC)
    #define FEATURE_VECTOR_SIZE FEATURE_DIM
#else
    #define FEATURE_VECTOR_SIZE (2 * ceil(sqrt(FEATURE_DIM)))
#endif

Run

./DT_MPC 0
./DT_MPC 1
...
./DT_MPC n-1

2 Secure Markov Inference

Constant.h

INFER_BATCH: batch size of SecMarkov

choosing different secget protocol by param USING_BASIC and USING_FOLD.

USING_BASIC	true	false	false
USING_FOLD	false	true	false
the secget protocol	Secget_Basic	Secget_Fold	Secget_Perm

• 2-gram

#define Markov_INFER_LOG_BATCH 1000
#define CP_DOMAIN_SIZE ((ALPHABET_SIZE*ALPHABET_SIZE) + 1)
#define IP_DOMAIN_SIZE (ALPHABET_SIZE)

• 3-gram

#define Markov_INFER_LOG_BATCH 10000
#define CP_DOMAIN_SIZE ((ALPHABET_SIZE*ALPHABET_SIZE * ALPHABET_SIZE) + 1)
#define IP_DOMAIN_SIZE (ALPHABET_SIZE*ALPHABET_SIZE)

run_Markov_eval.sh

• 2-gram: model_n="2"
• 3-gram: model_n="3"

./run_Markov_eval.sh 0
./run_Markov_eval.sh 1
...
./run_Markov_eval.sh n-1

Help

Any question, please contact [e-mail].

Contributor

Faculty: Prof. Weili Han

Students: Haoqi Wu (Graduate Student), Zifeng Jiang (Graduate Student), Wenqiang Ruan (Ph.D Candidate), Shuyu Chen (Graduate Student), Xinyu Tu (Graduate Student), Zhexuan Wang (Graduate Student), Lushan Song (Ph.D Candidate), Dingyi Tang (Post Graduate Student)