Decision Tree Training

Java/Kotlin implementation of the ORE-based decision tree training protocol. This implementation was developed using OpenJDK 11. Building and running the program will work with Java 11, but newer versions of Java might work as well.

Updatable ORE

The updatable Order Revealing Encryption Scheme is implemented in src/main/java/decisiontree/ore/updatable. It can be used by first specifying a Key-Homomorphic PRF, generating ORE parameters, generating a secret key and then encrypting/comparing:

var khPrf = new NPRPrf<>(NPRPrf.SHA256, Curve25519Group.INSTANCE, Prf.HMAC_SHA256);
var parameters = new UpdatableParams<>(khPrf, 8);
var sk = parameters.generateSecretKey(new SecureRandom());
var ct = sk.encrypt(2);
var m = sk.decrypt(ct);

Ciphertexts implement the Comparable interface, so Collections.sort() or similar will work.

Decision Tree Training secure protocol

The Decision Tree Training Protocol is implemented in src/main/java/decisiontree/main/MPCTraining.kt.

Running from gradle

To run the protocol with 50 training data points for each party, first run

./gradlew run --args="alice n=50 dataset=MNIST"

After the program has been compiled and is running, run

./gradlew run --args="bob n=50 dataset=MNIST"

for the second party to begin the protocol execution. The datasets will be downloaded automatically.

Building

To build the program run

./gradlew build

The built program will be in build/distributions/. To run the built program, untar the resulting tar file and run

bin/DecisionTree alice n=50 dataset=mnist

and once the program is running and waiting for a connection, run

bin/DecisionTree bob n=50 dataset=mnist